Asked Web Mania if they support
ASP.NET (C:\correspondence\WebMania)
My test ASP.NET website, so I don't lose it:
C:\Documents and Settings\user\My Documents\Visual Studio
2005\Projects\WebSite1\WebSite1.sln
Blogger: 1024MB
of storage available
My current website\final\devlog folder is 16MB
Gave blogger
(blogspot.com) a fair bit of testing, as a possible replacement for Excel for
the devlog.
Decided I didn't quite like it enough.IT'S TOO NARROW!It makes me
claustrophobic.
I do like the way it links from small preview versions of the images to the
full-size images; that's handy.
(below) I never
knew you could put flowcharts into Excel html pages, nice. Switching off
'autolayout' allows the nodes to be positioned freely.Might have been useful for planning the
menus.
Although, the font size keeps resetting and I can't seem to enlarge the
canvas.Also, the bubbles can become
too small for the text when I open the file in Internet Explorer.
C:\Program
Files\Le Cauchemar\
Profiles\
Default.txt
The game's default settings
test
test
Custom.txt
Player's custom settings
Excel: pictures,
flowchart nodes etc can be made into links - handy!
Web Mania's
support service was super quick, but the bad news is:
"As we have only Linux servers, hence you cannot
run theASP.NET application on our
server."
BOOOO!
123-Reg don't
support ASP.NET either:
"At present no shared 123-REG hosting packages
support asp or asp.net"
BOOOO!
Web Mania:
The most data I've ever transferred in a month is (June 2007) 1254.73
MB
Hosting is about £30/month inc vat
Bought 1 year of
ASP.NET-compatible hosting from Catalyst2.
Had a quick look at their control panel app, very nice.
22-Apr-2008
3.75
After putting the slave into
sleep mode while the game was running full-screen (master build) then
bringing it out of sleep mode:
hr failed, DXUT.cpp line 3484.
Tried getting at it with the debugger, but couldn't reach that line.
Tested MS Flight Sim demo; it handles this situation fine.
Added all the missing French
menu text.
Fixed various problems that
could arise from changing certain options using the menus.
(below) Replaced the various
anaglyph example profiles with a single one that should give the best
impression (minimal convergence, low saturation)
Example_TypicalAnaglyphGlasses.txt
#version 15.1
Released version 15.1
Just a couple of minor fixes and improved menu text,
nothing exciting.
21-Apr-2008
2.5
Asked the iZ3D people nicely if
they'd let me know how to support their displays. (C:\correspondence\iZ3D)
#revision 1088
ModeEnvironment.cpp: Fixed
crashes that would occur when starting the game with mist or falling rain
disabled (previously these class instances were only getting created if their
effect was enabled, and would therefore crash on CreateDeviceObjects).
#revision 1090
Fixed a bug with falling rain:
it had been using the wind's
speed to control the elongation of the streaks.So starting the game with wind disabled
caused almost-invisible streaks.
Also speeded up the rain falling speed a bit.Was previously looking a bit like neige
fondue.
I'm finding that
granddictionnaire.com is really helpful for translating these menus - it
contains a lot of very technical terminology (not surprising since it's
produced by the Office québecois de la langue française - those guys
ROCK!).
v15.1 will have improved French menu text.
TODO: fix ground glitch seen
when the game is started with rain impacts disabled
Still haven't been able to
reproduce (in debug) the original startup crash I was looking for (related to
failing assert in CModeStack::InsertAbove)
TODO: FIX THIS!
20-Apr-2008
Had a few people testing the
game today as a result of this flattering link (thanks crim!) :
To clarify, the game does not
have native support for iZ3D yet (I hadn't heard of iZ3D till today).
It does have native support for:
- Philips WOWvx 3D displays - http://WOWvx.com
- eMagin Z800 3DVisor (stereoscopy and 3DOF headtracking) - http://3DVisor.com
- NaturalPoint TrackIR with TrackClip PRO (6DOF) - http://www.naturalpoint.com/trackir/
- Anaglyph stereoscopy
More details in the release log:
http://programmerart.org/releaselog
and the FAQ:
http://programmerart.org/faq
I gather iZ3D takes in dual
monitor inputs (hurrah!), so that's the stereo mode I'll need to implement in
order to support these screens (which I definitely want to do, they sound
great).
So, sorry for any disappointment today, but do watch this space :)
Useful feedback from today's
testing:
- There's an intermittent startup
crash on v15.0 - this is top priority.Probably an
uninitialised variable, because it's never happened in debug.v15.1 will fix this crash.Test turning rain off, setting various
reses, etc.
- I should add a dual monitor stereo mode as quickly as possible.These screens could become very popular.
- The game is apparently compatible with iZ3D stereo drivers (so,
presumably with stereo drivers in general).
- It'd be worth me doing an optimisation pass on the pixel shaders about
now (framerate is unhealthy in full screen; in anaglyph mode it's
terrible).
- Options that require a restart need to make that fact clear (even if it's
just a bit of debug text in the corner of the screen).Eventually, none of the options will
require a restart, but till then it's important not confuse/annoy the
player.
- Screen res option needs to list just the enumerated available
resolutions; don't allow funny reses to be chosen, because people don't
expect them and don't want them.
...
...
- Some players didn't manage to find the text options profiles when they
looked for them (but ideally they should never have to).Maybe they were searching for *.ini ?
- Example anaglyph profile has way too much ghosting.Should definitely change this to very low
convergence and low saturation.Maybe
resurrect the anaglyph deghosting (would be a shame to waste that whole
weekend's work).
- Players seem to be impressed with the realism of the graphics (?!)
- It really is about time I added some weapon effects.This is more high-priority than the
enemies; people don't seem to mind the placeholder enemies too much.I don't think they're especially bothered
about sound either (although that will add tons to the experience).
- The videos on YouTube are useful.Need to add new ones; will probably do this once there's a weapon
effect to play with.
- People are actually playing the game!Aaaargh!!! Run away!
TODO: try asking them nicely if
they'll explain how it works.<-DONE
19-Apr-2008
0
Tried using Visual Studio 2005's
"Personal Web Site Starter Kit"
Managed to set up a holy grail of a website on a local server (brilliant
gallery and much more), but how do I then get that to run on my hosting
servers?
I don't see any mention of ASP.NET applications on the Web-Mania site.Am I just being thick, or do I need to
switch to one of the hosting companies that appear when I google for
("ASP.NET" hosting) ?
Had to re-install Microsoft SQL Server 2005 to get this stuff working
locally.
That's all I've managed to do
today.Spent about 4.5 hours on this,
which I've not logged because it hasn't produced anything usable yet.
I'm not going to persevere with
the new site or gallery until I've asked some people about ASP.NET.
Auditioned a few placeholder
music tracks.Zombie Overlords is
still probably my favourite.Tried
playing the game with that track combined with weather sounds - works well
and adds a lot.
TODO: placeholder music & weather loops, soon please!
Ooh, this track's nice,
especially the intro.I could really
use an intro like that.It's by
Cinematic Orchestra apparently:
http://fr.youtube.com/watch?v=7b8SAyRjid0
18-Apr-2008
0
Decided it was time I set up
some kind of screenshot gallery on the site.
Taking screenshots of ongoing work (and letting people see those screenshots) is a really good thing to do.It lets you get a feel for what's working
visually and forces you to fix any eyesores that are getting in the way.
Downloaded & installed PHP
5.2.5 (C:\installs\PHP).I don't know
or care what it is, but Gallery needs it.
Downloaded & installed
netpbm 10.27 (C:\installs\netpbm).Again, not the faintest idea what this is but Gallery needs it for
thumbnails to work.
...er, no wait, I don't want to
use Gallery because it requires me to set up my own server.
Downloaded and installed 'Web
Gallery Builder' (trial version), but it produces crumby unusable results.
I'm going to knock-up my own
gallery.Maybe Photoshop actions will
help for making the thumbnails?
16-Apr-2008
4.8
Added all the French text for la
foire aux questions.Took ages, but
not as long as it would have done in the past; I'm definitely speeding
up.
It's probably all wrong mind - please feel free to
correct my French!
Words that DON'T have cedillas!
:
- raccourci
- ici
- cela
- facile
- recevoir
Words that DO have cedillas!
:
- ça
- façon
- français
Took some screenshots from v15.0
(below) random v15.0
screenshot.Colour grading defaults
OFF in v15.0.Not because it's broken,
just because it's a counter-productive thing to have enabled until after I've
got the textures, lights and effects looking the way I want them.
Is there ever a good time to turn colour grading on?Maybe at the very end, by which point you don't need it any more because you've
improved the textures and lighting underneath.It's not something the game should rely on
in order to look good.
15-Apr-2008
4
FAQ = foire aux questions (nf)
website: added une foire aux
questions (programmerart.org/faq), because I'm gradually starting to need one
(which is good).
TODO: French text; the lack of this gives a really bad impression at the
moment :/
TODO: talk about anaglyph deghosting.
Spooky - I typed
"devlog" (pages francophones) into google.fr, jumped arbitrarily to
page 10, and my devlog was
right there!
HTML: "align" and
"valign" can't be assigned values inside a style string.Took me a while to work that out.
Found
that the eMagin SDK is now a freely-available download :)You used to have to do a little dance to
get hold of it.
Wait...that must mean none of that stuff is confidential any more (?)That would make it easier for me to share
my code.
14-Apr-2008
4
TODO?: for sharp [parts of]
shadows, threshold the soft shadow image instead of using the aliased raw
shadow image?
This would allow for a continuous change of softness along a length range
of the shadow (instead of the current cross-fade from sharp to soft).
Virtual-window headtracking now
only affects FOV if the 'auto-FOV' option is true.
TODO?: to reduce the extent to
which colour grading exaggerates banding, do the grading inside
HDRLighting.fx (tone-map the floating-point lit scene image, run it through
the ramp, then output it to fixed-point final render target.
The current banding is appearing because the lit scene image is quantized
to 8 bits per channel before
going through the ramp.
Found that the HDR\finalMultiplier option would cause
the rain and the edge blur to appear too dark if it was increased, or too
light if it was decreased (ie. those post-HDR effects don't take
finalMultiplier into account).
For v15.0, I've just disabled finalMultiplier and toneMap ("adaptive
luminance").It would be a shame
to throw these options out; instead what I'll probably do is change
finalMultiplier (todo: rename) to be applied at the same point where the
tone-mapping is/would be done.In that
way the post-HDR effects (and there will be more of them) won't have to worry
about it.Maybe the multiplier should
only take effect when tone mapping is switched off, or maybe I'll be able to
get rid of it completely.
TODO?: set finalMultiplier to 1 and tweak the post-HDR effects and the
middle-grey accordingly?
TODO: fix whatever that failing
post-build command is in Master config.
Made the Master config
remote-launchable.
Cleared-out some unused configs.
TOFIX: alt-tabbing away from the
game in fullscreen mode doesn't put it into pause mode when it should.
TOFIX: the south wall of Regent
St. disappears on v15.0.Looks like a
CullPlane problem, but what has changed?
Check the obvious first: check that the plane hasn't got shifted/rotated
(almost seems to be flipped).
Bit annoying, that, but I'm not going to let it hold up the build.
MAX: Added a hidden
'NaughtyStep' layer, and put one of the John St. road patches in there
because its normals were all wrong.TODO: investigate.Couldn't see
anything wrong with the decal's orientation.
MAX: For some reason every time
I tried to hide the NaughtyStep layer, the game failed the "is the
ground there" collision assert (?)
TODO: investigate!
For now, I've just shoved that one decal to the bottom of the stack so it
can't be seen.
Last-minute crash on the slave,
trying to open controls config.For
v15.0, I've therefore had to disable the controls config option :(
Tested it on the master and it worked fine.
For both cases, the Noctambule folder was renamed beforehand.
#version 15.0
Released version 15.0
TOFIX: frame-sequential
eye-swapping button (RMB) doesn't do anything in pause mode.
13-Apr-2008
9.5
TODO: for v15:
- fix processMenus option (currently causes crash if
true)<-BODGED(option is disabled for
v15.0)
- fix headtracking calibration mode<-FIXED
- fingers crossed that the OptiTrack stuff can
be made to work on Vista...<-FIXED(just needed
to install the OptiTrack SDK on the slave)
- slight specular / POM / mipping discrepancy
(see screenshot mentioned below)<-FIXED(was
screen-res filtering in DeferredLightingApply.fx)
- missing patchwork decals on slave<-FIXED(was uninitialised decal vert RGBA getting through to
Patchwork.fx)
- been getting a problem where the app sometimes loses all its keyboard
input after losing and regaining focus (does this only happen in pause
mode?)
- Deal with crash trying to open controls
config in full-screen.<-BODGED(fixed the
disabling of the config menu item in fullscreen).
- Alt-enter, etc<-FIXED,PARTLY(I've
enabled it again but the menus read alt-enter as enter to toggle options etc,
TOFIX).
Website: Added a YouTube video
bar; improved front-page formatting etc.
BUG: switching motion blur off
changes the filtering on the ground (dimples?).
#revision 1073
Menu system now ensures that a
disabled menu item is not selected.
TODO: Cars should explode when
you shoot near their petrol tanks.That will be a game in itself; can also be used as a way of taking-out
multiple enemies at once (the old 'exploding barrels' trick).
A really impressive explosion effect would be time well spent,
absolutely.By changing the parameters
(size, colour, etc) it should give a lot of mileage.
TOFIX: alt-enter, in pause mode,
is received as menu input (enter).
Same goes for alt-tab but that's only a minor glitch.
#revision 1075
Fixed a bug where you couldn't
quit the game if OptiTrack was being used.The quit message was somehow getting swallowed-up by the OptiTrack
message loop (in COptiTrack::Update).I expect my fix isn't ideal (re-posting a WM_QUIT if one was found in
the message loop) but it does work.The other messages (eg. I checked WM_KILLFOCUS) seem to be getting
picked-up just fine, so I don't know why QUIT is any different (?).TODO: Investigate?
#revision 1076
Renamed CNoctambuleApp::Quit
-> RequestQuit.The name was misleading because it just
posts a quit message rather than shutting-down the app there & then.
Re-arranged
EHeadTrackingHardwareHint so that it matches the string order shown in the
combo.
Remember: when adding new input
semantics, need to update the string table!(EINPUTSEMANTIC_)
Had a problem where the mouse
cursor would appear in headtrack calibration mode (allowing the window to be
manipulated.
A bit of fiddling later, got the same problem when entering pause mode and
couldn't control the menus either.
Just had to delete C:\Program Files\Common
Files\DirectX\DirectInput\User Maps\*.ini as
mentioned in the comments in InputManager.cpp.
Found a couple of problems with
falling rain in WOWvx mode:
1. background mapping / positioning is wrong
2. it's using the Z buffer for the ordinary screen size instead of the left
half of the screen (can see Z buffer 'ghosts' etc).
To avoid these problems I've temporarily disabled falling rain in WOWvx
mode for v15.
#revision 1078
Head-tracking calibration mode
is currently only available when using OptiTrack.TODO: enable it for 3DVisor too.
Failed to get v15 out.
Getting a crash on the slave when entering pause mode.<-FIXED(I'd forgotten to add Menu.fxo to the
installer)
Also not convinced the rain is right....
Will definitely release v15 tomorrow night.<-FIXED,
DONE
12-Apr-2008
8.91667
missing decal problem:
- not a sorting bug.Removing the base decal doesn't reveal them.
- not a visibility testing bug.Forcing visibility test to always pass doesn't reveal them.
- missing: doubleYellowJohnEastNorth
- IT'S UNINITIALISED VERTEX RGBA ON THE DECALS
MAX: X EXPORTER: brackets in
object names get turned into underscores in the X file
MAX: Adding a VertexPaint
modifier onto an object prompts the object into including a
"MeshVertexColors" block in the X file, defining vertex RGBA
(floats).
Removing the VertexPaint then removes the "MeshVertexColors"
block.
#revision 1062
FIXED: The
fix I've now made for the missing decal problem is to remove vertex RGBA influence from Patchwork.fx.
When I want to use vertex RGBA for some decals later on (it will be
useful), I'll just make a variation of Patchwork.fx called maybe
PatchworkVertexRGBA.fx (see define VERTEX_RGBA in Patchwork.fx).
I might well make this same fix for the building materials (eg.
Parallax.fx).The current placeholder
buildings have already been manually fixed by setting vertex colours or
applying VertexPaint modifiers.
#revision 1063
Changed DeferredLightingApply.fx
to use point sampling for the screen-res images (sceneDiffuse,
diffuseLighting, specularLighting).This prevents smudginess on 8800 which can filter floating-point
textures.
#revision 1064
DeferredLighting.fx: reduced the
diffuse angular falloff range a bit - doing that makes it nice &
contrasty, wetter and rougher, less 'default'.
#revision 1065
Found and fixed a platform
difference with the ground reflections.In ground.fx, the reflection sampler had been set to use BORDER
addressing.The x1900 doesn't seem to
support this (?), so it was using clamp instead (which works well for the
reflections).The 8800 supports
BORDER, so in places where the reflection is bent by the likes of a curved
road surface, black gaps were appearing in the reflection at the edges of the
screen.
Now changed the sampler to use CLAMP.
Only found one other use of BORDER in Shaders.sln: RainSplash.fx.Now changed that to CLAMP as well.
MAX: REMEMBER: After editing lines, don't leave them in
sub-object (eg. vertex) editing mode!
This was twisting some of my road paint.
Made some quick road paint
improvements in Max.
TODO: "Debug image"
option (enumeration).eg. setting this
to "rain occlusion" would show the just the alpha channel of
ERENDERTARGET_RAINOCCLUSION (which is also _FALLBACKSHADOWREFLECTIONPOSITION).
Maybe for the position 'images', I'd show the values divided by 1500 or
something, so you can really see what's going on.For the HDR luminance images, I'd show the
'exp' of the values.
TODO: "Debug image
mode" option (enumeration) :
- scaled PIP
- partial PIP (same scale as screen image but only visible in the PIP
area)
- full screen
- half of screen?diagonal split?
#revision 1067
Fixed some flickering seen in
the rain occlusion.This was simple Z
fighting.
I'm now using a minimal depth bias in RainOcclusion.fx to fix this
(DEPTH_BIAS_TO_PREVENT_Z_FIGHTING).
OptiTrack on slave:
In COptiTrack::Init,
CoCreateInstance(__uuidof(NPCameraCollection)...
is failing with
hr = Class not registered
"CLASSNOTREG indicates that the thing you are trying to create isn't
listed in the registry" (in the CLSID folder)
FIXED: OptiTrack mode is now
working on the slave (Vista 32).All I had to do was install the OptiTrack
SDK on the slave.It's a free SDK that
anyone can download:
http://www.naturalpoint.com/optitrack/support/downloads.html
("OptiTrack SDK for V100, C120, FLEX:3 Cameras")
Installing the OptiTrack SDK
(1.1.031) on the slave, got harmless messages saying that the following files
may have been left over by a previous installation (although there have been
no previous installations) :
C:\Windows\inf\
oem4.inf
oem5.inf
oem9.inf
"IDirect3DDevice8::CreateImageSurface
was renamed CreateOffscreenPlainSurface"
IDirect3DDevice9::UpdateSurface:
"This method is similar to CopyRects in DirectX
8."
Un-commented-out the code for
fullscreen controls config, and fixed compile errors (it hasn't been compiled
since I moved to DX9!)
Trying to open controls config
in fullscreen mode:
in CInputDeviceManager::ConfigureDevices:
IDirectInput8::ConfigureDevices fails (and doesn't call my callback).Returns this error code:
DIERR_NOINTERFACE
The specified interface is not supported by the object
E_NOINTERFACE
No such interface supported
11-Apr-2008
2.95
HDR: since g_apTexToneMap[0]
matches between the machines, technique "SampleAvgLum" (PS
SampleLumInitial) is off the hook?
So is it technique "ResampleAvgLum" that's going wrong?<-YES
I think I can see the root of
the problem now, and it's on the master side (X1900XTX, XP)
Every second texel of every second row of the g_apTexToneMap textures is very low (but not black).Now I just need to work out why.
FIXED!
I've fixed the HDR brightness discrepancy.It was the DITHERING renderstate affecting
the HDR luminance surfaces on the master but not on the slave (why the
difference though?).It was getting
initialised TRUE in CNoctambuleApp::VirtualRestoreDeviceObjects.
It's now initialised FALSE in that function, with a "can't touch
this" comment.Dithering is
nothing but trouble - I've seen this kind of problem with render target
effects a bunch of times in the past.
Screenshot showing the damage being done:
C:\noctambule\debug\11_4_8_HDR_masterPC_dither_vs_nodither.psd
Focus-reloads (eg. FX files)
work fine when remote debugging.
HDR brightness
discrepancy:
On the slave:
- g_pTexAdaptedLuminanceCur
is too dark, maybe something like 1/50 what it should be.Same for
g_apTexToneMap[0]
- HDR_OnCreateDevice fails to create that texture as R32F; uses R16F
- g_bSupportsD16 = true
- g_bSupportsD32 = false
- g_bSupportsD24X8 = true
- g_pTexScene seems
to match the master
- g_pTexSceneScaled
seems to match the master
- g_apTexBloom[0]
seems to match the master
On the master:
- the same path through HDR_OnCreateDevice and C_HDR::RestoreDeviceObjects
TODO: lights should all have a
slight fluctuation to them.No reason
for it, would just look nice.Almost
get this from the mist already.
TODO: continue tracking-down the
brightness discrepancy, starting from g_apTexToneMap[0]
9-Apr-2008
2.58333
TOFIX: (LOW priority)
Monitor-dragging while in pause mode causes a couple of minor menu glitches
(circuitry disappears and text is the wrong colour till selected)
Now starting on compatibility
debugging.
Tried using Pix for
Windows.
It doesn't seem to be able to give me a GPU timeline like the Xenon one,
that's annoying.
It does show me textures and surfaces.
It doesn't seem to let me analyse a remote PC.
- leamrotatedsharp.png (aerial guide photo), in non-shipped builds, is
loaded as a 22MB texture with 11 mips...and a corresponding 17MB surface [and
one for each mip?]Should probably
stop that happening ;)
- 1K TGAs (with mip chain) take up 5,592,404 bytes each.Same for 1K PNGs.
- 1K G16R16F = 4,194,304 bytes each.NB each RT has a texture and a surface, each the same size.
- 512 = 1,398,100 bytes per texture and per surface
- largest vertex buffers are 1,200,960 bytes
Made a Debug folder
(C:\noctambule\debug) which will contain debugging grabs.
Speckly junk in the alpha channel of the HDR backbuffer?:
C:\Noctambule\Debug\9_4_8_alphaJunkFromPIX.psd
TODO: need an FX file for the
texture displayed by CModeDebug.This
will allow the HDR surfaces to be displayed at a sensible brightness so I can
hopefully find the source of the brightness discrepancy between the two
machines.WIll also allow individual
channels to be displayed, inverted, whatever.<-DONE(10th,
DebugShowTexture.fx)
8-Apr-2008
2.75
Too-many-text-prims
problem:
- there seem to be 6 prims too many being drawn for each menu.Test: when I draw 7 less prims, half of the
final character is chopped off of either the English or French (whichever's
the longest) version of any menu if the final menu item is
fixed-length.
<-NOW UN-DONE THIS CHANGE; SEE COMMENTS IN CMenu::CreateItemGeometry
HA! At the end of
CMenu::DeleteDeviceObjects, it was recursing through all the child menus,
calling...wait for it...
RenderItems!
Handy!Good old cut'n'paste.Debug DX managed to catch the invalid
drawprim that uncovered this problem.
Without entering pause
mode:
Debug DX:153 references on closing.This becomes 150 after one or more monitor drags.Alt-tabbing doesn't affect it (reloadOnWindowFocus is currently
false).
"D3DX: MEMORY LEAKS DETECTED: 1516
allocations unfreed (1543638 bytes)
D3DX: Set HKLM\Software\Microsoft\Direct3D\D3DXBreakOnAllocId=0x373 to
debug"
I can't get BreakOnAllocId (from DX settings)
to work.
Retail DX: 633 references on closing, 1516 allocations
Can't get into the 3DVisor
forums:
"The website declined to show this webpage
Most likely causes:
This website requires you to log in."
(below) Trying to follow a link to an old
thread:
"Unable to open <URL>.The Internet site reports that the item you
requested could not be found. (HTTP/1.0 404)"
The paranoiac in me immediately assumes I've
been banned for some reason :/
All I said was "this headset's good enough for Jehovah"!
When dragging between monitors,
there are currently no unreleased device objects (only when closing).
CNoctambuleApp::VirtualInvalidateDeviceObjects:
When exiting, all effect pointers are NULL.Normal?NO <-FINE,
already released
CRenderTarget::InvalidateDeviceObjects:
When exiting, CRenderTarget::s_pListHead is NULL.Normal?NO <-FINE, already released
That's interesting:
If I remove the CGameMode:: AllInvalidateDeviceObjects and
AllDeleteDeviceObjects from CNoctambuleApp::FinalCleanup, I get the same number of unreleased device
objects when quitting.
Moved deletion of
g_game.pRenderTargets from CModeEnvironment::OnRemovedFromStack to
CNoctambuleApp::FinalCleanup.
I thought this would fix some unreleased device objects, but it didn't.
#revision 1057
FIXED.Can now exit cleanly again, and drag
between monitors cleanly.PHEW!!!
I don't fully understand the fix - just tried it on a
hunch and it worked.It's just a
matter of ordering anyway - all the right code has been in there for
ages.
It was the addition of DXUTCleanup3DEnvironment at the top of
CNoctambuleApp::FinalCleanup.This
also required the removal of AllInvalidateDeviceObjects and
AllDeleteDeviceObjects from FinalCleanup.
This finally clears an annoying blockage for v15.Will start compatibility debugging for v15
in the next couple of days.
7-Apr-2008
1.7
FX: The following causes
D3DERR_UNSUPPORTEDTEXTUREFILTER:
"MINFILTER = NONE"
or
"MAGFILTER = NONE".
NONE is only valid for MIPFILTER.
Now fixed in RainOcclusion.fx and RainFalling.fx
CMenu::RenderItems: trying to
work out why the wrong number of prims is apparently being drawn (caused
error in Debug DX because too many text prims were being drawn, for the
number of verts)
TOFIX: since moving some menu
init to CreateDeviceObjects, the processMenus option makes the menus crash
when opening.
27-Mar-2008
1.16667
TODO: when debugging the
glitches in v15, try running the game using debug DX and 'break on D3D error'
ticked.
Patchwork.fx, Parallax.fx: Now
fixed:
MIPFILTER = ANISOTROPIC;
produced an error message in debug D3D:
D3D9 Helper: IDirect3DDevice9::SetSamplerState failed:
D3DERR_UNSUPPORTEDTEXTUREFILTER
EEK NOW I'M CONFUSED, even the minfilter is
complaining when set to anisotropic.What's going on?
26-Mar-2008
2.16667
Grr, can't can't call virtual
function overrides from within a base class destructor (because the subclass
part is already destroyed) - I'd forgotton about that.
Instead, in the CGameMode destructor I'm now asserting that the device
objects have been invalidated and deleted (indicated by the mode flags).Added CGameMode::PrepareToDelete which must therefore be called before deleting a mode.Overloading the delete operator did cross
my mind... *shudder*
25-Mar-2008
2.16667
...nearly there; just got some
memory/resource leaks now.I can't see
any mis-placed calls to CreateDeviceObjects or RestoreDeviceObjects.
23-Mar-2008
3.3
CGameMode now inherits from
CListItem.
Added static CGameMode methods <Create,Invalidate,Restore,Delete>DeviceObjects
Removed the CModeStack equivalents, because
these methods need to be called on all the game modes whether they're on the
stack or not.
Made some good progress with the
device-changing stuff (eg. for dragging the game between monitors).
The game does actually survive this now; just need to fix up the menus and
that should be it all working.Presumably
the remaining unreleased objects were to do with menus?
21-Mar-2008
7.9
Added 'to-fix' notes at the top
of options.cpp
TOFIX: OptiTrack\cameraMountedOnMonitor: when true,
this causes a camera/projection crash on startup<-FIXED(just
had to put a check around the OptiTrack virtual-window headtracking update in
ViewInfo.cpp)
In a master build, alt-entering
now edits the value of the 'display\fullscreen' option.
In this way, the game remembers the state that the player left the game in
and starts in this mode when the game is next run.This also keeps the 'full screen' checkbox
in sync with the current state. (CNoctambuleApp::OnToggleFullScreen)
Menu.fx no longer writes to the
Z buffer.Previously, the menus had
been wiping away the falling rain if the pause background was frozen.
On the first frame, FTIME is
about 22.That must've been the
loading time.<-I've now added a temporary bodge so that FTIME
returns 1/60 until the environment has done its first update.I'll be overhauling the time system real
soon to allow for replays, etc.
#revision 1041
Spent quite a long time fixing
the annoying first-frame bugs that could be seen if the game lost focus
before it finished loading.
Most importantly, got rid of the crash.Also tidied it right up so that you can switch away while it's
loading, come back to it after it's finished loading, and it'll be sitting in
pause mode with a correctly-rendered first-frame environment behind the
menu.
Ideally I think I need a 'Start game' option at the top of the Pause menu,
that turns into 'Resume game' when you re-enter pause mode.
Started trying to fix the crash
when dragging the game window from one monitor to the other.I think I know what's going on.
When toggling fullscreen, the D3D device is lost and then restored.Nowadays the game
handles this correctly.
But when dragging between monitors, the D3D device is destroyed and a new one is created.
The problem is that various parts of my game code assume that the device
won't be destroyed until the game is shut down.So they're creating device objects in their
constructors and releasing them in their destructors.That creation and release needs to be
separated out now into CreateDeviceObjects and DeleteDeviceObjects functions.
Added
CModeStack::CreateDeviceObjects.DeleteDeviceObjects is already there and in use.
Renamed all InitDeviceObjects functions CreateDeviceObjects
g_game.OneTimeSceneInit creates
the mode stack.
TODO: continue fixing the
monitor-dragging stuff (add a CreateDeviceObjects for landscape, patchwork,
etc....).Looks like it should be
straightforward.
17-Mar-2008
3.08333
The following debug options now
work:
processFonts
processMenus
reloadOnWindowFocus
pauseModeOnWindowFocus
allowFrozenPauseBackground
CMIRenderTarget now displays
render target dimensions.
Added CRenderTargetOptions::WriteToMenuString for this
TOFIX: Why is custom.txt not
working (loading) ?<- it was being overridden by
phil.txt
Managed to get all 22 menus in
and working and not tangled-up with each other (well just about; 'HDR' and
'motion blur' are still overlapping).
To fix the tangle I had to split up the 'interfaces' menu by adding 'input'
and 'output' sub-menus.I was
reluctant about this, but I couldn't see another way, and works out more
future-proof anyway.
If the menus were to reshape/resize based on contents, and this affected
tree structure, it would be precarious (adding new items would change the
tree structure).
So tired and lethargic today;
haven't got much done.It's because I
was out drinking Red Bull last night.What do they put in that stuff?
YouTubed a little video of some
rain (nothing new) :
http://www.youtube.com/watch?v=1c3x__pKUFg
higher-quality version:
http://lecauchemar.com/videos/rain.wmv
I might record a longer, better version of this
idea later on
Joined the forums at
WordReference.com.User name: z6po
YouTubed a video of the
menus:
http://www.youtube.com/watch?v=EjvyaLIOzBE
15-Mar-2008
8.666
Option: replaced HDR\curve with screenEffects\colourGrading which
is a clearer name
String table now contains the
full set of configurable options.
Also did a lot of translation
The chap I contacted on ICQ
about the mythical WINx3D SDK never replied.So I've now uninstalled ICQ (it's evil and I only installed it so I
could contact him).
If anyone reading this manages to find the WINx3D SDK, please PLEASE let me
know!
Maths: added function WrapInt, and WrapIntExclusive which has
non-inclusive limits
MICombo value selection is now
cyclical/looping
Got nearly all the menus in.
14-Mar-2008
2.5
Been reorganising the menus,
re-grouping the options in the string table, mostly to avoid the
enormous-graphics-menu problem that I had.
Also translating.
13-Mar-2008
2.25
Lost some time due to a silly
'scratchpad'-contention bug.
To prevent this happening again I've made g_game.generalTemporaryBuffer private and added methods LockGeneralTemporaryBuffer and UnlockGeneralTemporaryBuffer (which just use asserts to check that everything's ok).
MICombo now takes into account
all its possible variable text when calculating bounds.
Added CMenuItem::UpdateBoundsForVariableVerts for this.
TODO: same for the other item types (for spinners, check the space required
for the min and max values).
TODO: limit the left position for menu contents; scale the contents to fit
the remaining space where necessary?
12-Mar-2008
3.78333
Menus now draw correctly in
WOWvx mode.
Fixed bugs with CMICombo
variable text.
CMICombo is now functional.
Stereoscopy modes are now listed
in order of how difficult they make it to read the menu ;)
TODO: for spinners (and
sliders?), definitely need to be able to type the numbers in :/
This creates a bit of work for me.
Might well leave this till after v15.
Added function COptions::ApplyStereoscopy
Found that zero-convergence
anaglyph (or practically zero) is super-clean.Obviously it
minaturises the scene and only allows objects to come out from the screen rather than
having depth into the screen.But I
like it - it's much more usable and impressive than the accurate version
(which is ruined by ghosting).
TODO: use settings something like this in the anaglyph example
profile(s):
convergence: 0.1
saturation: 0.6 odd
viewing distance: 28 odd
About half of the current menu
items are functional now.You can
change settings and see the difference they make to the game.
Found that upping (say,
doubling) the rain impact density really makes the rain look more
dramatic.And it ties-in better with
the amount of falling rain.
TODO: increase the default,<-done,
(256 -> 512)
check the speed hit.
TODO: I'm at the point where I
could reeely do with some timers, now that I can adjust the graphics settings
easily.
But that's another post-v15 thing.
Menu items now take a real
brute-force approach to applying their option changes.They call COptions::ApplyAll, then they
force a refresh of the environment image.
I was finding that the 'apply' work required after making any given change
could be difficult to gather-together.For example, after changing stereoscopy settings,
CEnvironment::ApplyOptions needs to be called.
The brute-force method removes all the uncertainty.
My worry with it was that it could easily cause a dropped frame that would
swap the eyes in 3DVisor mode.TODO:
test this on the new PC.
...And then there's the business of changing render target options,
errgh!....
I really wish they'd designed the 3DVisor with two monitor inputs (I did
see a mod for this, but you had to remove the headtracker).
11-Mar-2008
4.41667
sprintf: to left-align an
integer, and set its minimum
number of characters, and prefix it with a space if it's signed &
positive:
"%_-6d" where the underscore is a space
There doesn't seem to be a way of specifying the maximum number of
characters
Replaced
CStringTable::StringLength with CStringTable::MaxStringLength which takes
into account all languages.
This is because I'm going to allow the language to be changed from the menu
(since I've got a nice multilingual stringtable system, I think it's worth
the effort to show it off).
Did some translation, for some
of the menu options added so far.It's
probably all wrong.I'm currently
calling anaglyph deghosting "défantomisation" (citation: http://abtutor.free.fr/ColorGhosts/index.html)
To add to font map:
ç (shouldn't need uppercase)
(
)
?
<-Done
Font: Made the changes above;
also shifted the most common 'variable' characters to the start (numbers,
etc).
Found that lens blur is much more tame than Gaussian blur (ie. it doesn't
spread out as much).I think I prefer
the nice soft look of Gaussian though.Current blur settings: Graphics\Menu\blurSettings.png
CMICombo now displays its
variable text correctly
CMISpinner and CMISlider are now
usable :)
below: used the menu to adjust realWorld \ viewingDistance as a
test; this gives a glimpse of what the menu tree actually looks like.As you can see, I'm not finished
translating yet ;)
Can't wait to see it with proper reflections on it, assuming that whole
thing works....
TODO: localise decimal points
into virgules.I'm assuming sprintf
won't do this automatically on a French-language system (??)Good thing I've taken the precaution of getting a French-language system so
I can test this kind of thing :D
10-Mar-2008
2.5
#revision 1025
Got CMICheck updating its text
the way it should, phew.
I was just being thick and forgetting to transform the text verts from menu
space to tree space.
CMISpinner is now displaying its
variable values correctly.
9-Mar-2008
2.31667
CMICheck is now closer to
updating its text the way it should.Not quite got it yet.
8-Mar-2008
3.3
Added function
CStringTable::StringLength
Added function
CMenuItem::PlotText
CMICheck now responds to input,
to toggle its bool value.
Full screen' menu option now
works.
CMICheck is now just-about
updating its text the way it should.
7-Mar-2008
1.43333
Late start again.
#revision 1021
TEMP: The 'controls' menu item
now disables in fullscreen mode to prevent the current crash.
TODO: fix properly after v15.
Added CNoctambuleApp::OnToggleFullScreen for this.
DXUT.cpp is including global.h & main.h for this - TEMP?
Started making CMICheck (the
yes/no menu item) functional
6-Mar-2008
1.38333
Was working late, hence the late
start.
Got the auto menu positioning
fixed, properly this time - it's all making perfect sense now and the tree is
forming just the way it should.
Embarrassingly, it was the very simplest final step of the thing that I was doing wrong - distributing
each menu's angle range amongst the child menus.All the more complicated stuff was working
fine (as of last night).
#backup
Backed-up the project.
5-Mar-2008
4.666
Added quaternion normalisation
into MatrixLerp33 after finding some cases where the menu movements went
crazy without it.
Found that dragging the game
window fully onto the second monitor causes an error (unreleased device
objects).TODO: fix.The DX sample apps handle it correctly.
Fixed the problems I was having
with the auto menu positioning, but ran into yet another
one.Had a look at it using the
3DVisor in case this made the problem clearer, but couldn't see anything
obvious.Currently the stereoscopy
menu thinks it has hardly any space around it for its child menus.I haven't been able to work out why yet,
but I expect it could be something to do with it being at the top 'pole' of
the tree.TODO:
FIX!<-FIXED(6th)
3DVisor: On the old PC, the
headtracker was working fine in the bottom two USBs, but for the top two I
had to install its drivers.Working
fine now though.
3DVisor: Got a 'lost connection'
error one time just after entering pause mode.Don't know why it happened (loose USB
cable?) but it was all handled gracefully.The game switched from full-screen back to Windows to show the
message, in the right language; the retry option succeeded; the game
reappeared full-screen and continued with headtracking.
This was using the 3DVisor in the second monitor port (along with a second
monitor).
Added the 'anaglyph' menu
4-Mar-2008
0
Works night out.
3-Mar-2008
2.75
Added the 'stereoscopy', 'real
world', 'display' and 'weather' menus.
CMISlider now inherits from
CMISpinner.I might rename these
classes later to make the relationship more obvious.
Option
display\headMounted is now realWorld\headMountedDisplay
The 'display' menu is more for the properties of
the video output; the 'real world' menu contains all the real-world info.
Made some fixes & added some
tests to the automatic menu positioning.Ended up running into another bug after adding
more menus.TODO: FIX<-done(5th)
2-Mar-2008
8.55
#revision 1014
Menu highlight effect has
changed in design slightly.Was
originally a solid-edged bar with alpha-blended red fluid going through it
(see mockup 17th feb), representing the blood of the menu.The functional aim was to thicken/darken
the background for the text.
Now it's a subtle additive
red smoke, with no solid edges, drifting from the letters as if they're
subliming.Can be a wee bit hard to
read the text when a light is behind this additive smoke plus the text glow,
but for the purposes my demo, prettiness wins over practicality.And it does look very pretty now.
Added method
CMenuItem::Enable.Menu items can now
be disabled, which makes them un-selectable, removes their text glow and
turns the text to roughly the same weak colour as the menu titles.
As I add new items to the menus I'll disable any that aren't
hooked-up/supported yet.
F1 is now another pause key.
Decided that I don't want to
support direct-from-game access into the DX controls config.Partly because the time doesn't freeze
properly at the moment when doing this.
I'm keeping the 'CONTROLS' input semantic because some controllers actually
have a 'controls' button.For my game,
this will open up the Controls menu (my one).
Temp: 'Controls' link now simply
opens the DX controls config.
The menu tree now automatically
generates the positions of all the menus.Currently still a bit wonky; todo: fix.<-sorted(5th)
1-Mar-2008
10.6667
Added method g_game.Quit to exit the app.Made
sure this doesn't bypass any DXUT error checking (message boxes about
unreleased resources, etc).
Added option debug\reloadOnWindowFocus (reload shaders etc when the window regains focus.In Master this behaviour is always false).
Added option debug\pauseModeOnWindowFocus (enter pause mode when the window loses focus.In Master this behaviour is always true).
CModeStack::Remove no longer asserts that the mode is currently on the stack;
bails out instead (see comments).
Realised that I needed to let
DXUT handle at least some hotkeys (eg. alt-enter); changed it back
accordingly.
Also realised that DISCL_EXCLUSIVE wasn't what I wanted because it blocks
the above and blocks the printscreen key.Now using DISCL_NOWINKEY instead to disable the Windows key.
#revision 1006
DXUT.cpp: disabled DXUT handling
of hotkeys:
VK_F3 (toggle reference
device; currently I don't support this, so it crashes)
VK_F8 (wireframe; I
don't support this, so it's glitchy.Can be handy now & again but I'll re-enable it manually when I
want it)TODO?: A well-implemented wireframe option on the debug menu would
be very handy for showing-off culling, patchworks, effect geometry, etc.Would be handy for debugging too.A good thing to add, really.
VK_ESCAPE (exit app;
this key now enters & exits pause mode instead)
Added option debug\allowFrozenPauseBackground (allow g_game.flags.freezeBackgroundWhenPaused to be set true
where appropriate).In Master this
behaviour is always true
#revision 1007
Fixed the rain-splash glitch
seen when alt-entering while paused.
Added flag g_game.flags.refreshEnvironmentImages which allows the 'frozen' paused environment to re-render for
a single frame after alt-entering.
Related to this, HDR no longer pauses when the game is paused.It causes less problems this way (eg.
alt-enter), and also it would make no sense for it to be paused when
headtracking's enabled because you can then still look around in pause mode.
Crosshair no longer appears in
pause mode.
Menu tree now aligns with the
current camera (so the menus appear in the right place).
#revision 1008
Selected menu items now glow
white (see CMenu::UpdateMenuItemVerts).The glow fades out in
sync with the circuit line.
Photoshop: found that tiling marble textures are the easiest
thing in the world to make:
- difference clouds, times lots, 15 say
- curves
- done!
FX: Crazy: In menuCircuit.fx (and presumably any
other effect file) I have to set ZENABLE = TRUE in order for the thing to
*write* to the Z buffer - despite the fact that there's a separate
ZWRITEENABLE which I also have to set TRUE in order for the thing to write to the Z
buffer!
Got a blue-screen crash while in
Dev Studio doing nothing in particular.
Got the menu selection
'highlights' working nicely.Took a
bit of experimentation to get them playing well aesthetically with the other
menu elements (especially the circuit lines).
29-Feb-2008
2.71667
Menu.fx: Added a quick &
dirty refraction effect, and improved the placeholder reflections to take
into account the tree orientation.
TODO: get the default hotkeys
available again in debug builds;
clean-up the game exit (from pause mode).Is there a DXUT exit function??<-done (see 1st Mar)
28-Feb-2008
2.83333
#revision 1002
Got a rare crash when drawing
lighting on the first frame, while the game window was in the background (I
was doing a little E&C build at the time, hmm).Crashed because g_game.pCurrentCamera was
NULL.
Had a look and saw that the pointer wouldn't be set until after the
lighting would draw (???)
Now fixed.
Menu transitions no longer give
any movement glitches with a low framerate.
I've also smoothed-over a movement glitch that could be seen when a new
transition to a third menu was started before a transition between two others
had finished: the homer-interpolated transition matrix is now used as a lerp
target for the final tree matrix used for drawing.
Added function MatrixEqualApprox33 - this is used
to detect the menu being at rest at the end of a transition.
Added flag g_game.flags.freezeBackgroundWhenPaused.When the game is using
no headtracking and no stereoscopy, this flag is set true so that in pause
mode, the last rendered scene image is used instead of re-rendering the same
scene each frame (note: the rain is still rendered each frame because it's
done post-HDR).
The result of this is that no matter what framerate the player is getting
for the game, the menus should always run in a frame if stereo &
headtracking are inactive.In such a
case, it makes up nicely for the lack of motion blur on the menus.
Seeing the menus moving against
their background makes it clear that they need some refraction to stop them
looking paper-thin (or hollow like balloons).The refraction has to correctly follow the branches too, otherwise the
illusion is broken, so I'll need to use the per-pixel normals.<-DONE(29th)
TODO: add refraction.<-DONE(29th)
Also improve placeholder reflection to take
into account tree orientation<-DONE(29th)
TODO: on entering pause mode,
the tree needs to orient itself so the root menu is on the line of sight (not
including the head-tracking offset).<-DONE(1st mar)
27-Feb-2008
4.5
Got the menu transitions working
nicely.
Added CHomer (Homer.cpp/h) which automates smooth movements between two
float values.It even lets you change
direction smoothly half-way through a transition between menus.
It's surprising how getting little things like the menu movements just
right creates an impression of quality.The menus of a game are kinda like the interior of a car in that way.If you want the player to feel like they're
driving a Jaguar, you shouldn't use the menus from a Škoda Favorit.
26-Feb-2008
1
DIUtil.cpp: Now setting
cooperative level to DISCL_EXCLUSIVE.This disables the Windows key.<-(see 1st Mar)
Windows key can't be blocked
during control config.The player
can't map it to to an action either.
DX docs:
"Even if the cooperative level for the
application is disabling the Windows logo key passively through an exclusive
cooperative level or actively through use of the DISCL_NOWINKEY flag, that
key will be active while the default action mapping UI is displayed."
Input: Couldn't see a way to
prevent the likes of DIKEYBOARD_WEBSEARCH doing its default thing (although
detecting the button is no problem).
I can't block it in LowLevelKeyboardProc (dxut.cpp) unless I know its VK_
code, and I can't see one that corresponds.
Input: DIKEYBOARD_RMENU is 'Alt
Gr'
DIKEYBOARD_APPS ("Application" in the controls config) is the
'menu' key (normally for right-click menus)
25-Feb-2008
1
I've set DXUT to no longer
handle default hotkeys (esc, pause, wireframe, ref device).<-(see
1st Mar)
I've also set it not to parse the command line (just to be safe, until such
time as I actually want it to).
Did a few input/menu odds &
ends.CMILink now changes the current
menu.
Revived my old MySpace page
(which was still being linked to from the band page) to point people to my
site.Hopefully it won't use up loads
of my time like it did before.I'm currently
keeping it very minimalist.
24-Feb-2008
12.25
Had a look at the
LocalDeformablePRT DX sample:
C:\Documents and Settings\user\My Documents\Visual Studio
Projects\LocalDeformablePRT
Had to increase the shader model numbers in
CDXUTDirectionWidget::StaticOnCreateDevice to get it to run, because 1_1 is
no longer supported.
Also had a look at the PRT demo.
Didn't see anything very useful-looking for this project, but it was
interesting as reference...
Input: A DIACTION array, for a
DIACTIONFORMAT struct, for IDirectInputDevice8::SetActionMap, is not allowed
to contain two mappings of the same input (dwSemantic member).
Nor is it allowed to contain out-of-genre inputs (eg.
DIBUTTON_BROWSER_PREVIOUS can only be used if the genre of the DIACTIONFORMAT
is DIVIRTUAL_BROWSER_CONTROL).For
this reason I allow DIBUTTON_FPS_PAUSE to enter pause mode and DIBUTTON_BROWSER_PAUSE to exit pause mode.In practice
though, I can't test either of these 'virtual inputs' so there's also a
non-virtual input to do the same thing (defaulting to P).
Finally worked out how to get
the menu input channels working.I
just have to call CInputDeviceManager::SetActionFormat to switch between reading game actions (ie. during play) and
reading menu actions (ie. when paused).I'll probably also use this mechanism for fly-cam, headtrack
calibration, replay mode, etc.
#revision 992
Input 'action' descriptions
(seen when configuring/viewing controls) are now localised through the string
table.
See EINPUTSEMANTIC_... strings.
Added OnChangeOfLanguage methods to a couple of classes.
Input: semantics mapped to
DIAXIS_FPS_MOVE don't appear in the config interface, and aren't active, when
using a keyboard & mouse.I'm
guessing it would show up if I used an analogue joystick/gamepad.
Managed to get controls
configurable [again?].Just had to
supply a name string for CInputDeviceManager::Create.
Control files in C:\Program Files\Common
Files\DirectX\DirectInput\User Maps now get named
accordingly (upper-cased).
Spent most of the day giving
CInputManager a thorough overhaul.It
had been based on the input code from an old DX aerial FPS sample (no longer
in the SDK).There was a lot of lazy
duplicated code there so it was bulky and hard to maintain.
Now it's tons better, nice and tidy and makes perfect sense.Added CAxisInput, and CButtonInput (which
takes care of debouncing).
Aarrgh!Just noticed that CInputManager was written
four years ago !Que cela me serve de leçon.
F1 in fullscreen: assertion
failure at C:\noctambule\source\common\diutil.cpp(344): hr hr=E_INVALIDARG
(0x80070057)
I've only just discovered fr.wiktionary.org - it's ace!
TODO: need an analogue
joystick/gamepad thing to test some of these control axes.
According to Wikipedia, an
"invert mouse" option means flight-sim mouse, and apparently Quake
set that standard.I was never sure
which way round it was but I'll go with that.
Added option controls\invertMouse (defaults false)
NOTE: this option isn't limited to mouse input; it inverts the final
accumulated pitch value (from all devices) used for looking up & down
during play (but doesn't affect head-tracking).Does this inside the input manager.
23-Feb-2008
2.75
Faffing with the input
code.I've mostly made sense of it
now.
21-Feb-2008
5
FX: I can't get the BLENDOP
state to do anything!It's always just
adding!(???)
(below) Got the circuit lines
finished.They were sodding fiddly!I didn't make it any easier for myself by
insisting that they have perfectly rounded
corners (which are too small to even notice
anyway).Ah well, it's all
experience.
You can cycle through the menu items of the active menu now.The lines fade out pleasingly when they
disappear.
TODO: get those menu input channels working.
20-Feb-2008
2.38333
Menus: Made more progress with
the circuit lines and general functionality.
19-Feb-2008
2.18333
Menus: Made good progress with
the circuit lines and general functionality.
Enemies: Sketched an
almost-upright scorpion beetle guy with head-pincers, various legs (longer at
the front, for uprightness) and lobstery head-fangles/feelers.The body is fat enough that he'd be a
satisfying target that could explode when hit.The scoprion tail emits a snappy lighting
attack.
(I'm not including an image here because it's not a great drawing [good by
my standards though!] and don't have a scanner anyway.)
I'll try sketching a few different designs in the near future and then
decide which one I like best.
Would be nice if the enemies had sort-of searchlights on/in their head,
like miners' lamps, maybe glowworm-coloured to contrast with electric-blue
attacks (an idea which I'm keen on).If they have fangles & feelers, they would cast groovy shadows and
could demonstrate backlighting/scattering.Fangles would very likely be code-drawn, and code-animated to provide
nice secondary motion.
18-Feb-2008
3.16667
Enemies: Spent about an hour
collecting images of lobsters, toads, scorpions, spiders and other ugly
animals & insects.
The toads have good faces and skin but they all look like they could be
reasoned with.Maybe they'd be more
intimidating if they were snarly (like angry monkeys - sharp teeth) instead
of looking bored all the time.
The scorpions look the most intimidating.
Made some progress with the
circuit lines for the menu selection effect.
17-Feb-2008
8.16667
#revision 977
(below) got the menu normals
about as good as they need to be.I'll
save any further improvements till after the reflections have been added.
Made a placeholder sphere map by
chopping-up some old location photography : ETEXTURE_SPHEREMAP
This'll do for the menus until the proper reflection effect is written (see
7th Feb).
#revision 979
(above) Menus now look like
this.
The smudgy effect at the edges of the screen (screenEffects\edgeBlur) is a new addition in v15.I like it because it's dreamy; it also kinda suggests water in the
eyes, which is good.
If I wanted a stronger impression of water in the eyes, I could make the
blur thicker at the top of the screen.I might consider that later.In
3DVisor mode, the effect is disabled by default.
(above) Mockups for the
selection effect (mockups\menus\selection_PCB.psd)
- The text lights up white
- A PCB-diagram-style line connects from the empty space on the left to
underline the text (maybe the whole line, maybe just the label).Blend mode: soft light.
- A background box with faded left & right sides contains clouds of
colour scrolling from left to right.I've previewed the movement in Photoshop - looks great, very organic.In red especially, it's as if it's the
blood of the menu.Blend mode: normal.
TODO: should the 'Link' menu
items also have PCB lines to their target menus?That might be nice; do a mockup.
Added the WOWvx menu.
16-Feb-2008
0
Band practice
15-Feb-2008
3.58333
(below) menu normals are close
to being what they should be.The
texture is still rough though :
14-Feb-2008
4.11667
menus: Had to make fixes to the
branch positioning, because I had got it slightly wrong.The branches are now drawing nicely.
TODO: for the joins to work properly, need to take
into account the length of the branch (map the branch V at constant
world-space density)<-FIXED(15th); did it in a
different way
Had a look at the menus in VR
mode - they're good :)
This was also the first time I've seen the falling rain, paused, in VR
mode.It's quite spooky.I'm definitely glad I took such a
stereo-friendly approach to the the rain.
13-Feb-2008
3.33333
Did a bit of experimentation
with detail mapping for Parallax.fx.Didn't get anything worth bothering with.
See define DETAIL_MAPPING_TEST.
TOFIX: option 'general\autoFOV'
seems to have no effect<-WORKS FINE NOW(21.3.8)
Got the menu branch join points
located properly; last night's stuff worked well and only needed a few little
fixes.
12-Feb-2008
4
Mapped the texture onto the menu
ellipses, and improved the texture so that the branches will adjoin neatly.
D3DXCreateTextureFromFileEx, in
CTexture::Create, seems to be forcing square/POW2 textures, despite the
D3DX_DEFAULT_NONPOW2 flag (?)
This was causing a problem for the menu texture, but I've now made it POW2
dimensions (as it should have been, of course).
Drafted the code that locates
the join points for connecting the branches onto the menu ellipses.
This code also calculates how far to sink the branch into the ellipse so
that the texture will join up seamlessly.
The whole of tomorrow night will be spent getting this working properly.
11-Feb-2008
3.16667
Got the menu transformation
matrices working properly.
Added function Vec3RotateX
Planned the menu branches - they
should join on quite easily.
Made a normal map for the menus (menuNormals.tga /
ETEXTURE_MENUNORMALS).From menu.psd
Added the Graphics and Debug
menus.
10-Feb-2008
11.0833
Added a 'Remote Debug' build
config.<-removed, no longer needed (just using Debug
and switching the debugger type)
Finally managed to find the USB
driver (or driver installer) for my modem:
C:\Installs\BT Broadband\Util\BT220V.exe
Both the PCs are now talking to each other and to that
t'Internet.
This puts me at the point I should have been at four days and 55 pounds ago
(the router I bought, which I couldn't get to work properly, is completely
redundant now).
Added a 'Remote Local Debug'
build config (debug the remote exe locally)<-removed,
no longer needed, now using local exe for remote debugging
NOTE: the debug exe, should I
ever want to include it in the install, requires d3dx9d_34.dll
These dlls can either go into the system folder
(where the installer currently puts them, maybe shouldn't), or just next to
the exe.
The debug exe is about 3 megs
Vista: HANDY: Start \
eventvwr.exe \ Journaux Windows
Shows details of recent crashes etc.
#revision 953
KER-REMOTE-DEBUG!!!POW!!:D
Now, I am the master.Er, literally, yeah.
And I'm getting D3DX debug
output in French, kewl.
remote debugging: had to set the
'Remote Command' to be the
slave-relative path to the exe ( \\master\c\... ).(I'd forgotten this stuff because it's been
8 years since I've used remote debugging.)
The game exe and the remote-debug monitor are both kept on the master
(keeps things simpler & tidier, and seems to be the approach recommended
in the msvsmon docs).Does mean the
game takes even longer
to start up, but that needs fixing anyway.
Had to set the 'Remote Server Name' to be the server name displayed in msvsmon (currently
PC-DE-PHIL:4015 ).
Had to set 'Debugger Type' to 'Native Only', and 'Connection' to '...no authentication' because I'm using Home Editions of
Windows on both PCs.....Corresponding
settings had be chosen in msvsmon.
Grrr:
"In Visual Studio 2005, edit and continue is not
supported when debugging native C++ code remotely"
http://connect.microsoft.com/VisualStudio/feedback/ViewFeedback.aspx?FeedbackID=101862
It's not supported in Visual Studio 2008
either:
http://forums.microsoft.com/MSDN/ShowPost.aspx?PostID=2274488&SiteID=1
Just to see what would happen, I
tried enabling authentication for my remote debugging.Got the message I'd expect:
"Unable to start debugging.
Access is denied. This seems to be because the 'Network access: Sharing and
security model for local accounts' security policy does not allow users to
authenticate as themselves. Please use the 'Local Security Settings'
administration tool on the local computer to configure this
option."
The option it's referring to is not configurable on Home Editions
of Windows, so there will no remote EnC for me.It's annoying, but it's no different from
using a console devkit, so I should stop whining.When I want to use EnC, I'll just debug
locally.
Installed NVIDIA PerfKit 5.1 on
the new PC.
Can't get NVPerfHUD to work though; apparently there isn't an instrumented
driver for 8800 yet:
"The NVIDIA Setup program could not locate any
drivers that are compatible with your current hardware.Setup will now exit."
Got the remote debug monitor
auto-starting with all the right parameters, so the game can now be launched
with remote debugging without me having to do anything.
Installed NVIDIA PerfKit 5.1 on
the new PC.
Can't get NVPerfHUD to work though; apparently there isn't an instrumented
driver for 8800 yet:
"The NVIDIA Setup program could not locate any
drivers that are compatible with your current hardware."
I just got confused to hell and
back - Visual Studio had set itself to 'show all files' on the solution
explorers.So I was opening the
shaders solution and thinking "what is all this weird random old junk?Where are all my .fx files?"The thing that made it extra-confusing was
that no amount of SVN reversion seemed to change anything.
The problem might have been caused by me typing on the wrong keyboard,
which I've been doing all day - although there doesn't seem to be a default
shortcut for 'show all files'.
#revision
956
#revision 958
Very nearly lost the following
files, which weren't in SVN, while heavy-handedly messing about trying to
understand the problem described above:
h_Menu.fx
h_MotionBlur.fx
MenuTitle.fx
CullPlaneUser.cpp
Luckily I found them in the recycle bin.These files are now in SVN.
This is why I back-up my actual project folder, rather than the SVN
repository.
Should really back-up both I suppose.The repository is only 1.5 GB.
Visual Studio: lost my keyboard
shortcuts (?)<-FIXED (11th) somehow
'MyMacros.vsmacros' had got unloaded
9-Feb-2008
2
Wasted the whole day trying and
failing to set up
- a router
- remote debugging
Project settings: I've set the
debugger type to 'native only'.This
is because remote debugging on home editions of Windows only supports native code rather than
'managed' code.I don't even know what
this means, but I wanted to check that the debugger still behaves ok in that
mode.Seems fine so far.
Project settings: Resources:
I've set 'culture' to French (France) (0x40c).I don't think this makes any difference to
my stuff anyway.
#revision 951
Deleted the 'media' folder (junk
inherited from the HDR sample).
#revision 952
Deleted the 'UI' folder and
removed it from the installer (junk inherited from the HDR sample).
Cleared out a load of inherited
UI junk from the HDR code, including HDR_InitApp.
7-Feb-2008
3.66667
The need for reflections on the
menus led me into thinking about reflections in general.The most important place I need them is on
all the glass (shop fronts, etc), but I also want the cars to be reflective,
and plenty of other stuff too.
I've come up with a very devious plan which, at the lowest detail setting,
works without redrawing *anything* :
1. Ignore the ground reflections
for just now; I'll keep them working the way they are.At the end, they might well join up with
the general reflection
system described below; that would be great, but it's not vital.
2. There will be a 'number of reflection planes'
option, kinda like the numProjectors one for the shadows.It will be possible to set the number as low as 0.Each of these reflection planes will do
pretty much what the ground reflection is already doing - straightforward
planar reflection.They'll get
assigned to the wall planes that need them most at any given time.I'll be aiming to set the default number of
reflection planes to 1 (ground only) or maybe 2 if I'm feeling generous.
3. Here's where it starts to get
novel.I will be keeping a very small
list of nearby planes (walls and ground) that will be used for simple ray-tracing.I might well use the CullPlanes for this,
since they'll already be positioned to coarsely cover great big swathes of
wall, which is ideal.One of the
planes will always be a big section of the ground plane around the viewpoint.
4. One offscreen render target
will contain a collage of images roughly representing the surfaces of each of
the planes.Those images are created
by, each frame or across a span of frames, the appropriate trapezoids being
cut out of the previous frame image (possibly also out of the ground
reflection image), reshaped and glued onto the ORT.Obviously only little bits of the collage
(the bits that were in the frustum) will thereby get updated on any given
frame, but that's fine.
When a new wall plane becomes 'active' as a raytracing plane, if it's not
in view I can start it off by slapping a 'generic wall' texture on it or copy
from whichever wall plane has been active longest, something like that.
5. The collage from point 4 is
used for the reflectors that don't have a reflection plane (so this includes
all non-planar reflectors).The
following also applies to those reflectors.
In the reflection receiver material, whether it's on a window, a car or
whatever, the vertex shader will trace the reflected view ray to the point
where it hits a raytracing plane (if it doesn't hit a raytracing plane, bail
out ASAP).
An extension here, to support normal mapping, would be that the vertex
shader computes an extra two intersection points - one for the ray reflected
by the tangent and one for the ray reflected by the binormal (or by vectors
somewhere between those axes and the normal).The pixel shader would then interpolate between the texture coords
calculated in point 7 for each of these intersections.
6. The vertex shader can compare
the reflected ray direction against the main view direction, and against the
view direction reflected by the ground plane.From this it can decide whether the previous frame, or the current
ground reflection image, would the better image to use for the reflection.
7. The intersection point is
transformed into the clip space of either the previous frame's view, or of
the current ground reflection view - whichever was chosen in point 6.If the intersection point is found to be inside
the frustum of interest, simply sample from the image chosen in point 6, and
we're done.
If the intersection point is found to be outside the chosen frustum, sample
from the collage instead.Cross-blend
at the transitions between the frustum and the collage if necessary - it
won't break the bank.
It's going to be a lot of work,
but the result should be really pretty amazing.Provided it all works,
and doesn't look pants, the benefit-to-cost ratio of this effect will be
totally off the scale :)
Although it's very tempting to
dive into the reflection work now, I really need to get v15 out first.This is because the
compatibility bug in v14 (missing ground detail) is annoying, and slightly
worrying too.
So the menus in v15 will not have any
reflections on them at all,<-added placeholder
reflections for now (17th)
but I will hopefully get the menus functional
(most of them at least), so that options can be tweaked on the fly, finally -
yipee!
TODO: mock-up and implement
menuitem selection effect<-done(17th) (mockups\menus\selection_PCB.psd)
6-Feb-2008
3
Got the new PC working
nicely.And - PHEW - the 3DVisor
headtracking is working on Vista 32 :)
Greatly improved the default
settings for 3DVisor stereoscopy.
Remembered that I was clamping stereo convergence (0..1).It's best to allow this to be set higher
than 1, so I've now removed the clamp.That should let me push the background further into the distance
without changing the other settings.
The setting that surprised me
the most was interpupillaryDistance.Once I'd got a decent
impression of depth going on and the objects were appearing at the right
size, it still felt like they were lacking 'weight' somehow.Reducing the
interpupillary distance from 7 to 5 fixed this really effectively.Suddenly the cars felt as big as real
cars.
These three stereo parameters - field of view, convergence and
interpupillary distance - balancing them properly makes all the
difference.It's the difference
between "yeah, that's kinda 3D" (the worst feedback a VR project could ever receive)
and "it feels like I'm actually there".
Couldn't manage to network the
two PCs using this broadband router.Neither of them seems to be able to talk to it using USB.
TODO: GET A ROUTER!
I managed to get TrackIR working
on the new PC, but for some reason the game isn't finding the camera.I did remember to copy the 'OptiTrack'
folder into the 'NaturalPoint' folder.
Remote debugging might give me more information.I hope it's fixable and not some kind of
limitation with the OptiTrack SDK?
TODO: need to try keeping the
two eyes in sync (only updating once per two eyes); they lag apart on
v14.Or have I already changed that
now?
5-Feb-2008
2.41667
Hah!I just got a crash because the 'O' in my
character table was actually a zero! D'oh
below: got the fish-eye and
subtle diffusion working for the titles.There's a slow, just-noticeable bulging motion of the title text, as
if it's breathing.I'll probably do something
similar for the tree itself; it's a good way of making the thing seem
'organic'.
The next layer is some glossy
reflections, which should make it feel really solid and tie it into the
environment.
But first I'm going to set up
Vista 32 on the new PC...
4-Feb-2008
4.5
Added MenuTitle.fx - Effect for menu title text.WIll be closely related to MenuItem.fx
below: more soft-lighty
goodness.Pretty-much got these menu
titles working now; just haven't quite nailed the fish-eye distortion.
Need to fix the glows as well; currently they get a bit chopped at the
edges (faint vertical lines in picture below).
Google: I've set France as the
'geographic target' for programmerart.org and lecauchemar.com.I'll be interested to see if this brings in
any more French hits.The hit rate for
the site is rubbish at the moment.Why
is no-one visiting my site?!
3-Feb-2008
7.5
CLockableTexture can now be
given a callback and context to be used at the end of OnResetDevice.
CFont instances use this to measure their characters, when option debug\processFonts is true.
Added file FontCharacterData.hpp - contains the array of character data #included into Font.cpp
(I'm currently only supporting one font, but that's easy to change later if I
need to).
This is auto-generated by CFont::MeasureCharacters when option debug\processFonts is true.
Font blur: Gaussian, 30 pixels
(for 1024 font image) .... I think
TODO?: rain bounce mist effect
by blending the blurred scene onto all the transitions between top surfaces
and farther-away not-top surfaces?
below: Got all the font stuff
working well.This is built from the
ground up, not using any D3DX stuff.The blend mode is a cut-price forgery of Photoshop's 'Soft Light'
blend, performed by MenuItem.fx's pixel shader.C'est joli hein?
I'm following the mockup pretty-much to the letter, and it's paying off.
TODO: Menu
titles <-done (5th)
and selection effect.
2-Feb-2008
9.16667
Added an error dialogue to
handle unplugging of the 3DVisor while the game is running:
MESSAGE_3DVisorError_removed
TODO: need to zero the headtrack transform if
the player chooses to continue without headtracking.
TODO: looks like I need to do some 3DVisor re-initialisation if the player
chooses to retry.
Excel: ctrl+` to show all forumulae 'raw' rather than evaluated (discovered
this by accident because it's my Visual Studio 'find in files' shortcut)
Added functions:
Vec2Zero
Vec2InitMinBound
(initialise a minimum-bounds vector to high positive values)
Vec2InitMaxBound
(initialise a maximum-bounds vector to low negative values)
Added options:
debug\processFonts -
(TODO) measure the characters on the font texture(s) and save out the
measurements<-DONE(3rd)
debug\processMeshes
- (TODO) optimise all meshes and save out the optimised versions (this will
reduce the game's startup time enormously - currently it optimises the meshes
each time it starts)
TODO?:
D3DXCreateTextureFromResource ??
#revision 933
Spent a long time extending
CLockableTexture so that it can load itself from a texture file (this is so I
can lock the font map to measure the characters, when option debug\processFonts is true).
Had a lot of ordering & resource-handling hassles to get my head round,
most of which were caused by the complete deletion & reconstruction of
the pause menu tree each time the pause mode is entered (intentional, debug
only, for ease of testing).
But it's working now, at last :)
TODO: Measure the characters<-DONE(3rd)
1-Feb-2008
below: a petition to free our
friend from a horrible
situation in Dubai; you might be able to help him by signing it:
Made a decent chunk of progress
with the font & menu stuff.I'm on
the brink of seeing some letters on the screen now :)
Added MenuItem.fx (effect
for CMenuItem text)
Remembered there is no
D3DPT_QUADLIST >:(
30-Jan-2008
3.5
Improved the font texture
loads.It's now square as well.The glow seems to fit ok; will need to wait
& see if it's wide enough for what I want (although I'm not likely to want
to change the layout now!)
Added CFont, which takes care of
calculating basic positions and UVs for text verts (CFont::PlotText).
It owns the character table and each CFont instance owns a texture.In practice there will only ever be one
font instance, but I'll write it as if there could be any number of them.
CFont: Decided that I probably
don't want to get into the business of automatic line-wrapping, and I
shouldn't need to.
27-Jan-2008
5
The online devlog is now the
complete devlog from day 1.This is
mostly to try to draw more hits in.
Internet Explorer seems to deal pretty well with huge pages, eg. doesn't
try to download pictures until you scroll to them.
Tried using the GDI function
'GetGlyphIndices'.Supposedly I would
just have to include <windows.h>, but this didn't help; the call still
didn't compile.Not sure what I'm
doing wrong there.
Tried using Photoshop's
File\Automate\Web photo gallery option (after running the action called
Select Workspace "Web Design")
Output here: C:\Noctambule\Website\gallery
I'll try a different template next time.
Note: ImageReady lets you save
out SWF files.I don't know how to
make them animate, link or any of that stuff though.
Installer: Discovered the right
way to ship fonts for the game: just set the font file's 'register' property to 'vsdrfFont'.
When you install the game, it magically installs the font! :)
And it uninstalls the font when you uninstall the game (but presumably not
if it's marked 'permanent').
15*8 = 120
8*16=128
Diacritiques utilisés en
français :
http://fr.wikipedia.org/wiki/Diacritiques_utilis%C3%A9s_en_fran%C3%A7ais
Eventually went off the idea of
using D3DX font stuff.It was *almost*
helpful, but not quite.I got the
feeling that it was going to make things more complicated than they needed to
be, and less flexible.
Built-up the font map PSD
(C:\noctambule\Graphics\Fonts\Menu\font.psd).It took a while but it's looking sturdy now.I'm leaving plenty of room round the
characters so I can have a blurred (glow) version that uses the same
coordinates.
See corresponding string array currently at the top of menu.cpp
26-Jan-2008
10.25
Added file pairs for the menu
system (most of them are stubs at the moment) :
PauseMenuTree - subclass of
CMenuTree for the pause menu tree
Menu - a menu, a
node in a menu tree
MenuItem - base
class for all menu item types (below)
MILink - item that
navigates to another menu
MIConfirm - eg. for
the 'quit' option; pops up a confirmation dialogue.Might use this for the undo options
too.
MICheck - checkbox,
for Booleans.This'll be the most-used
menu item type.
MISpinner - for int
and float values
MISlider - eg. will
be used for 0..1 type values
MIRenderTarget -
render target dimensions (and in some cases, format)
MICombo - for enum
values.Slightly slangy use of the
Windows term 'combo', since there's no 'edit box' element to it.
Note the absence of a basic
'text box' base class, and of any mention of 'boxes' in the naming.No boxes here.
All menu item types have a string index property, because they'll all
display at least one string.
Added ESTRING_none for uninitialised strings.
Strings: started using these
naming conventions for menu strings in the string table:
(ESTRING_)menuName_title -
the menu title string
(ESTRING_)menuName_optionName - the option's label string
(ESTRING_)menuName_optionName_confirm - the
confirmation message for an MIConfirm item
#revision 913
StringTableParser: fixed a
parsing bug that was stopping some entries getting written out.
#revision 914
...and another one.
C++: Weird - it seems that a
constructor can't directly call
a pure virtual method of its class (doesn't link), but it CAN call such a
method via another (non-pure-virtual) method.
eg. CMenuTree::CMenuTree can't call CMenuTree::Populate,
but it can call CMenuTree::Init, which calls CMenuTree::Populate<-no, that gives a runtime 'pure virtual function call'
error.
I'm now calling CMenuTree::Init from CPauseMenuTree constructor instead;
this works fine.
Note: CLinkedList is not
circular.
Added CLinkedList::DeleteAllElements.
#revision 916
Made some fixes to
CLinkedList.Spent a while getting
frustrated at some memory trampling problems I was getting with it.Fixed now.
#revision 917
I fell-out with
CLinkedList/CLinkedListItem and have now deleted them
(they still exist in SVN history).
I've replaced them with a slimmer, wiser CListItem base class, which is nice
and minimal (ListItem.h)
C++: It's a shame you can't pass
parameters to destructors:
error C2524: a destructor must have a 'void' parameter list
#revision 919
Another stringtable parsing fix
(all the French text was coming out English)
Started finding out about D3DX's
font stuff.Looks useful.
'Text3D' sample:
C:\Documents and Settings\user\My Documents\Visual Studio Projects\Text3D
Some info about optimised
anaglyphs:
www.3dtv.at/Knowhow/AnaglyphComparison_en.aspx
25-Jan-2008
0
C's leaving doo (another one off
to France!)
EB recommended Monpellier
again.I'll very likely visit there
this year.
NC says the game wasn't running
too well on an 8600.
24-Jan-2008
3.25
CModeHeadTrackOffset, when added
to the mode stack, now makes the hardware-specific profile
'currently-editing'.This'll save the
player a load of confusion.Returns to
the previous profile when removed from the stack.
This means the player can never graphically edit
headOffsetMatrix in the Custom profile (by design).
below: Got an accidental
3rd-person mode, by setting a large head position offset when in 3DVisor
mode.This created a head-tracked
pivot cam, which is really quite nice to use!So I'm tempted to add a mode like this:
Didn't really get any work done
this evening.Preparing and uploading
a video takes hours :(And it leaves me in a distracted mood so I
can't get any coding done afterwards.
On the up-side, I know how to
use Windows Movie Maker now ("ga-ga!"), and I know its output works
with YouTube.
I used the Video for local playback (2.1 Mbps) setting.
23-Jan-2008
2
New PC: Just needed to send some
power to the CPU.Now working fine
:)
Just waiting for Vista to arrive now.
Added function
CModeStack::InsertAbove
CGameMode now inherits from
CNameUser
CModeStack: fixed the ordering
within the various add & remove methods so that the mode is fully
added/removed before OnAddedToStack/OnRemovedFromStack is called.This makes it safe for modes to add/remove
child modes when they're added/removed.
Added function
CNoctambuleApp::SetPauseMode
22-Jan-2008
0
Built the new PC with tons of
help from M.At
the end of the night it powered-up but didn't do anything.I assumed I'd broken the CPU or the
motherboard.
21-Jan-2008
0
Started trying to build the new
PC.Used the wrong screws when trying
to fit the motherboard (misled by some instructions).
Need pliers to continue.Can't find
pliers.
TODO: find pliers.
20-Jan-2008
10.5
Increased
CNoctambuleApp::s_generalTemporaryBufferSize from 2K to 16K.
CNoctambuleApp::(g_game.)generalTemporaryBuffer is used to buffer the file
contents in COptions::Save.
It's also used by CBlockFileParser and CRainFalling.
(These were previously called MAIN_TEMPORARY_BUFFER_BYTES
and g_temporaryBuffer)
#revision 901
COptions::Save now saves out
every option correctly.
Moved:
COptions::ReadRenderTargetOptions
is now
CRenderTargetOptions::ReadFromFileString
CBlockParser: All the 'Read...'
methods are now static.The likes of
CRenderTargetOptions uses these functions; I'm happier to expose them than to
force classes to inherit from CBlockParser just to get at them (which could
be confusing if their job isn't to parse blocks).
Fittingly, all these methods were already marked as const-this anyway.
CBlockParser::ReadVec3/Vec2/Matrix
now take pointers instead of references - more consistent with the rest of
the code and therefore less likely to cause mistakes.
I reckon this music track would
be brilliant for the replay mode:
The suitably-titled "Zombie Overlords" by DeathBoy:
http://deathboy.anti-goth.com/ToRights/zombie%20overlords%20-%20DeathBoy.mp3
It fits well because it's dark, techy, intense,
glitchy and ominous-sounding.
TODO: choreograph the camerawork/cuts to tie-in
the with whatever track I end up using.
Options: rather than having
ok/cancel options, I'll add a button to revert to (re-load) the last saved
options.This'll need to reset the
override flags as well.
This option should be called something like "Undo last
changes"
Saving will be automatic on exiting the pause mode.
A separate option for "Restore defaults" will be provided
(empties the currently-editing profile, clears its override flags, re-loads
the profiles).
Undo: If changes have been made
in the current pause session, the undo option above needs to change to 'undo
current changes' (reload all current files - doesn't require a backup).
After undoing current changes, it should change back to 'undo last changes'
(copy from backup instance)
More undo: Each menu needs its
own local 'undo' and 'defaults' option as well.
For the partial profiles, I'll
need to put a choice in the (probably main) menu for whether we're editing
the Custom, OptiTrack or 3DVisor options (defaulting to Custom).
In debug builds, maybe include the option to edit the 'Default' profile as
well.
This setting will simply control which override flags gets set when editing
any options.It should probably spring
back to Custom each time you enter the pause mode, to save confusion.
Partial profiles: need some
visual indication of which options are overridden in the current profile, and
a way of un-overriding them (right-click?)
The override/unoverride option should toggle the value between its
unoverridden* value and its last overridden value (if any) from the current pause session.
*not necessarily default
Added a Profiles\AutoBackups folder which will allow the 'Undo last changes' option to
work (game-generated copies of the previous versions of the custom profiles
will be kept here)<-changed the undo system to
work without backup files now
Installer: Added
AutoBackups folder (ships empty)<-removed now;
not using files for the undo system.
Renamed 'Optitrack' folder to 'OptiTrack' - shouldn't cause any disruption
I've got a slight mix of
enumeration styles in the game.Nearly
all of them are old-style (as specified in my coding
conventions doc) :
like "ETHREADPRIORITY_LOWEST"
but three of them are EB-influenced and make use of scope (including one I
added just now) :
like "COptions::EEditableProfile::custom"
Either style could technically be used in the options profiles, but for
that case I'll definitely stick to the old style for consistency.
For the rest of the code.....well the scoped ones make the other ones look
bad.Scoped ones are better I'd say;
non-scoped ones are more popular/common; it's good to show consistency; it
can also be good to show flexibility.There doesn't seem to be an ideal solution without knowing the style
preferences of the target audience.
I'll just carry on using a mixture.However, I
won't wrap any enums in their own little struct/class unless I actually have
a need to.<-(see below)
Updated the coding conventions doc
accordingly.See also: log entry on
2nd of March 2007.
Hmm, now I see the reason for
wrapping those enums in their own little structs - it's because the
enumerator doesn't create proper scope for its contents the way a struct
does.
ie. Two enums in the same scope can't have the same member names.
#revision 906
Cool, the partial profiles
loading & saving is now working properly (saving a profile only
writes-out the values that the profile overrides).
TODO: stop it writing out empty category blocks<-done
CModeHeadTrackOffset (headtrack
calibration mode) now detects the player's changes and flags the relevant
option as being overridden, causing the new values to write to the
currently-editable profile when pause mode is exited :D
Where did I put that biscuit tin?
Well that's actually all the
options stuff done for now - a weekend well spent.This leaves me totally ready to start
putting the menu system in place, made all the more satisfying by being able
to hook into a *working* options system right from the start :)
19-Jan-2008
8.75
NOTE: strncpy_s appends a null terminator, in addition to copying the
requested number of characters.
#revision 899
I've renamed my #included CPP
files to be HPP files instead.HPP
seems to be the most correct extension for the job, according to the
description below, and it saves me having to exclude them manually from the
build when they're added to the project.
- OptionDescriptions.hpp
- StringsFrançais.hpp
- StringsEnglish.hpp
Updated the tools accordingly
HPP:
"Header file that may contain variables, constants, functions, and other
code referenced by a C++Builder [etc.] source code file; allows common
functions to be referenced by multiple files..."
Weird, I was sure there was a 'strcatf' ?
NOTE: the option description
system doesn't allow for private members in the option category structs.Not ideal, but I only had one of them
anyway.
#revision 900
Spent today writing the OptionsParser tool, which is
working well.COptions::Save now does just that
:)
Non-loaded members of option
category structs use the NON_LOADED symbol after their type, which is a
signal to OptionsParser to skip to the next member.
18-Jan-2008
2.8
PC is scheduled to arrive on
Monday.
Nailed-down the workings of the
"option descriptions" system from last night, using a compiling
mockup of the auto-generated code.This looks like it's going to work well.
NOTE there is still no automation for menu population, association of menu
items with options, etc.I don't want
to automate any of that, never did.
I've designed-in some support
for editing of partial profiles.SOptionDescription::flags is used
to indicate which profiles override their inherited value for the option (the
relevant profiles being Custom, 3DVisor, OptiTrack).
Added function COptions::Save, which takes a
filename and a filter for the flags described above.
Added file OptionDescriptions.cpp which will be auto-generated by the 'OptionsParser' tool.
TODO: OptionsParser tool!<-DONE(19th)
17-Jan-2008
4.25
#revision 881
Yeah, no point doing anything
with OptiTrack at the moment - I'll wait till the new PC's put together, and
see how that copes with it.
Till then, since I've made a bit of a mess and broken a couple of things,
I'm reverting COptiTrack.cpp&h to revision 876.
Added website\final\code to SVN.
OptiTrack: The weird startup
lerp was caused by a faulty matrix copy (like *destMat =
*srcMat where neither is a pointer, so it was only
copying the first float).
That's a relief - I'd imagined it was some kind of sinister, thread-related
hoy.
CModeHeadTrackOffset: Made this
much more usable by changing the transform control to be in head space rather
than TrackClip/headset space.
OPTIONS:
I need to do a bit of an overhaul in order to make the options nicely
saveable.
I could do this using fancy C++, wrapping each type in
a class so I'd have like COptionsFloat with a float() operator and so
on.
Instead I'm thinking of writing a little tool that parses Options.h and for
each category builds up a CRC-sorted table of names, CRCs, types,
(strings&counts) and member pointers.And the same kind of thing for the categories themselves.Something like the following (but obviously made to compile) :
OptiTrack thread:
- had to raise the priority to above-normal to get it to move at all in
fullscreen.
- when windowed and paused, it's reasonably smooth.<-when windowed & running and set to
THREAD_PRIORITY_TIME_CRITICAL, it's very good
- when running and/or
fullscreen, it's jerky.
Tried running the TrackIR heads
in the background while the game was running in a ~1280*1024 window:
- it's jerky just like my in-game tracker thread
- when its priority is set to 'realtime', it copes better and uses as much
as 20% CPU
I might leave this stuff till I
can test it on the new PC.I feel like
I'm wasting my time at the moment.
I've tried everything I can think of; it's as if there isn't enough CPU
speed for the image processing to happen fast enough - which maybe suggests
that it's not done in the camera hardware itself like I'd imagined?Not sure.
15-Jan-2008
3.5
Added option debug\disableEnemies (replaces a macro called ZOMBIEMANAGER_NOTHENOO ;)
I'm trying to keep all the 'debug' properties named so that their defaults
are false (hence 'disableEnemies=false' rather than 'enemies=true')
Remembered that the fallback
shadow image is generated on a small ORT, so the zbuffer can't be used when generating it.
If I definitely only want
fallback shadows on the ground (or shall we say, the 'reflective mesh'....)
then FallbackShadowReflectionPosition.fx can be merged into
FallbackShadow.fx, and the fallback shadow image can be generated by
rendering the reflective receiver mesh with this combined effect.
<-nah, that would mean drawing the receiver mesh for
each light.It's good that currently
the receiver only has to be drawn once (RT covering quads from there on) to
generate the images for all the lights.
Got the fallback shadows working
properly.
FallbackShadowReflectionPosition.fx: removed the constant 'cMainViewDir'
The depth occlusion now works by comparing VP-relative squared scene distance with squared view-space
receiver Z position.(NOTE the slight mismatch; if it
causes any visible problem then I'll fix it - so far it looks fine).
FallbackShadowReflectionPosition.fx: removed the constant 'cFallbackShadowsOnGroundOnly' (now
always true).
Using fallback shadows on meshes other than the intended receiver produces
glitchy, unreliable results (works well in some cases, certainly, but that's
not good enough).
TODO: option for ground (kerbs)
to reflect in ground (backface culling should help a lot here)
This will also produce kerb fallback shadows, which will be nice.
#revision 876
Realised I could rename
StringsFrancais.cpp to StringsFrançais.cpp, and did so (purely to be
awkward), using SVN rename.
Updated the StringTableParser tool accordingly.
Threads: found that without the
Sleep(0) sub-loop in each of the thread procedure loops, Windows locks up.
TODO: finish converting
COptiTrack to use a thread to read the camera.
14-Jan-2008
4
Trying to track down this weird
reflection / fallback shadow glitch:
- rainOcclusion = false: doesn't change it
- numProjectors = 0: doesn't change it
- remming-out the fallback shadow drawprim doesn't change it
- ERENDERTARGET_FALLBACK_SHADOW looks fine
- ERENDERTARGET_REFLECTEDDEFERREDLIGHTING_DIFFUSE looks fine
- remming-out the CRainImpact drawprim doesn't change it
- writing height=(1 or 0) from parallax.fx doesn't change it
- writing height=(1 or 0) from ground.fx doesn't change it
- writing gloss=(1 or 0) from parallax.fx doesn't change it
...ERENDERTARGET_REFLECTION_FINAL
#revision 870
Fixed most of the problem I was
seeing (corruption / shadow after-images in the ground reflection
image).
DeferredLighting.fx was
reading shadow samplers even for the reflected draw (in which case they were
just whichever textures were last used on those two stages).
I've now added uniform parameters to toggle shadow work and to toggle the
'diffuse shadow' behaviour (fallback shadows set this false to only shadow
specular lighting).
This fix should be a fairly hefty savings too :)
Fallback shadow masking is still misbehaving (and always has been).That's why specular has been missing on
walls etc (showing up only occasionally, with faulty masking corresponding to
the ground behind the building).
- FallbackShadow.fx has a bit of code to prevent shadows when the light is
behind the viewpoint.It seemed to be
causing far more glitches than it prevented, so for now I've remmed it out.TODO: test
properly once the other glitches are fixed.
- When the ground mesh is drawn into the fallback shadow 'reflection
position' image, the landscape seems to have already been cleared out of the
zbuffer.The landscape absolutely must be in the zbuffer at that
point in order for the masking to work.TODO: reorder stuff so that this is the
case.
- clearing ERENDERTARGET_FALLBACKSHADOWREFLECTIONPOSITION to a specific
value and then testing for that value just doesn't seem to work.There's must be some catch I don't know
about, probably specific to floating-point render targets.
#revision 871
Right, I've got something partly working now (still got the
sorting problems), by using 0.f as the 'no shadow' value (tested-for in
FallbackShadow.fx).TODO: investigate/improve, etc.
See CLandscape::GenerateFallbackShadowReflectionPositionImage.
13-Jan-2008
0
Spent today catching up with
work-work.
VirtualDub: a good setting seems
to be Indeo 5.10, 50% quality, force keyframe every
frame (30fps).
Below: dancing head video with those settings,
youtube-ified.Also synced the music
which was 1 beat behind :
The crazy background thing last
night was caused by an uninitialised W column in
CHeadTracker::vsHeadTransform.
Added adaptive smoothing for
TrackIR - works a treat; filters off all the jittering.
Removed option headTracking\smoothing
Added option OptiTrack\smoothingRange (the total object pixel distance above which the lerp T is 1)
Added a little unit test for the
COptiTrack's vector tracking (generates artificial objects to track) - COptiTrack::SpoofObjects.I really should have done this to begin
with - it took about 10 minutes and it's helping the debugging
enormously.
By generating artificial objects, I'm removing the following sources of
inaccuracy:
- IR camera FOV measurements
- TrackClip light position measurements
- IR camera resolution
Using the artificial objects,
the tracking (especially looking at depth results) is bang-on except when the clip is at certain 'ambiguous' angles, then it
shakes briefly same as usual.This is
reassuring - it narrows down [most of] the problem to
COptiTrack::InterpretObjects.In fact,
the jitter will almost certainly be coming from the ray-plane intersection in
GetDistanceFromRayToRing.I would love to know how a mathematician
would solve all this.....
Tried using a different
implementation of IntersectRaySphere (from t'Internet again).Currently sitting alongside it and named
'IntersectRaySphere2'.Compared the
difference in return value (inside COptiTrack::TestZPosition); it never
exceeded 0.001f.
Both versions gave the occasional jitter described above.
TrackIR 4 camera res:
355*290
355/290 = 1.2241379310344827586206896551724
#revision 869
Wow...just been playing with
virtual-window headtracking + full-strength anaglyph + placeholder weather
sounds.
It's a spicy combo!
TODO: maybe add an
anaglyph-friendly (drier) version of the lighting?Specular highlights on the ground seem to
be the main things making the anaglyph really ugly at the moment.
Did another bout of testing
& calibration for TrackIR.....
- I'm now using the officially quoted horizontal FOV
of 46 degrees, *exactly*.
- The aspect ratio that works in conjunction
with that FOV give the correct distances for the clip is exactly 6:5 (1.2:1),
so that's what I'm now using.
It's possible that instead, I should be using
the aspect ratio of the camera res, adjusting the fov to fit
accordingly.Who knows.It seems to be working well anyway.
TODO: a fix is needed in the VW
headtracking: I'm not convinced the camera correctly moves closer to objects
as the head moves forward; investigate!
Parallax.fx: made the top sides
and undersides of objects automatically wetter
TODO: really should try a
rainbow ramp texture for the volume lighting.
Added option OptiTrack\cameraMountedOnMonitor (defaults true)
This indicates that the OptiTrack camera is on the real-world screen
plane.
When this is true and a camera is active, realWorld\distanceToScreen is
ignored and the head depth percieved by the camera is used instead.
Noticed that alt-entering while
paused leaves the rain (and presumably bloom?) over-bright (alt-entering
again while not paused fixes it).
Could this be a clue to the over-brightness bug on
nVidia cards???
Had another look at the TrackIR
control panel app.Their head-tracking
results are extremely solid and precise compared to mine.
One of the reasons why it's so good seems to be the clever smoothing that
they're using - the interpolation rate seems to be weighted by the delta
between the current and destination transforms/dots.In this way, pixel-level jittering is
completely hidden, while large (genuine, intentional) movements feel
perfectly responsive.
Added a disclaimer in the
release notes, and in the public headtracking source, saying that my TrackIR
support is a bit naff.
Added function Vec2Distance
Got the virtual-window fov
working; currently just sorting out some smoothing stuff.
#revision 864
Below: an uninitialised matrix
or something started causing this crazy random flickering background.So I made an impromptu music video - enjoy!
Little detour this evening - I'm
going to add the 'virtual window' head-tracking mode, having seen a video of
it looking very effective using a Wii-mote.
When using OptiTrack, this mode shouldn't suffer so badly from the accuracy
problems I'd get normally, because it's only the head's position that we care about, not its orientation.
Removed
CCamera::ApplyProjection.This work is
now done in CViewInfo::ApplyCamera.
FIXME!!! HDR\finalMultiplier
breaks the brightness of the rain and the other effects using the bloom
image.
Now writing HEIGHTMAP values to
position.a (MRT[1].a)
DeferredLighting.fx: Added
heightmap-based fake self-shadowing for diffuse.Works lovely.
#revision 860
DeferredLighting.fx: Tried doing
something similar for specular.Decided it wasn't worth it (it worked, but in practice you hardly get
to see it), and it damaged the glinting of the rain impacts too.
DeferredLighting.fx is highly speed-critical.
DeferredLighting.fx: Ambient is
now affected by height.Works lovely.
Ground.fx: Reflection strength
is now increased at areas where the reflected object is very close to the
ground.This makes ground-to-landscape
joins look more natural.
#revision 861
Very pleased with all the
improvements to the lighting model.
The POM is looking mighty chunky now, even on the ground :)
7-Jan-2008
3.75
Improvements to texture levels
(road, kerbstones, brick) and Ground.fx.
Ground.fx now has nice Fresnel-esque stuff
IMPORTANT: Keep colour grading
turned OFF while tweaking texture levels!!
TODO: finish fixing Ground.fx
(see #errors - to do with reflection position)<-FIXED(8th)
TODO: really need to streamline
the texture tweaking process.Maybe
use that procedural T-pages idea (model them in Max and make the game
assemble them) ?
TODO: decide what to do about
kerb decals (overhang)
6-Jan-2008
11.3333
Colour grading (HDR\curve) is
now forced OFF in anaglyph modes.This
is because:
- grading increases contrast; contrast is bad for anaglyph
- grading affects colour balance, which wants to be kept flat for
anaglyph
- grading is expensive; anaglyph is expensive enough already
- grading was causing a rain colour mismatch in anaglyph modes
NOTE: bloom source image is too
banded to be of much use on its own
3DVisor: roll drift seems worse
than I remember it, hmm.I'm not
getting any yaw or pitch drift though.
Lack of yaw drift would seem to suggest that the problem isn't magnetic.
Added option motionBlur\secondaryBlur (defaults true / false for 3DVisor).
This blends the scene towards the bloom image according to each pixel's
speed (done in HDRLighting.fx).
This hides the gaps in the motion blur that appear when turning
sharply.
When using an HMD, the benefit of this option is negligable and it should
not be used.
Added option screenEffects\edgeBlur (defaults true / false for 3DVisor).
This blends the scene towards the bloom image at the edges of the screen.
#revision 847
Getting an instant-restart crash
whenever I try to draw to MRT[3] when drawing the menu.
- tried setting the MRT[3] target on MRT[1]; that works fine
- tried a different target as MRT[3]; still crashes
Moving the menu draw to inside CPatchwork::Render prevented the problem
(see revision)
Found that the menus will need
to draw after the bloom image is created, otherwise they feed back and become
opaque.
Drawing them after the bloom image is created means they can't have any
motion blur or deferred lighting.
Motion blur on the menus was really nice while it lasted, but you can't
have everything.
#revision 851
Started playing about with
texture brightness levels, starting with the road surface - which now looks
about the way it should (inky black with a just a scattering of grey
speckles).
TODO: Ground.fx / DeferredLightingApply.fx: texture
colour has to get even darker in the distance (shallower angles) - so that
road paint appears to blend in with the tarmac.<-DONE(7th)
5-Jan-2008
10.8333
RenderSubset: NEW TEST: now only
generates rain lighting for the first of the two eyes.<-no
dedicated rain lighting image now, see below
TODO: VR mode: need
to improve eye swapping method so that the ordering is never wrong (swap
using a signal to the headset).
Renamed CMist::Generate -> CMist::Prep because nowadays it
doesn't do any rendering, just prep (it's rendered as part of
DeferredLightingApply.fx)
#revision 833
Moved pRainFalling->GenerateLighting to be done straight after CHDR::EndScene (instead of right
before the rain render).
The lighting image no longer lags behind by 1 frame when used on the
MenuTree.<-see below
CGameMode::flags is now public
#revision 835
Made important changes to
falling rain:
- Did-away with rain lighting image.Rain now refracts the HDR bloom image (which I've set-up to be an
unbiased blurred version of the scene).
- HDR bloom and glare are now drawn in a separate pass after falling rain
has drawn.
- Falling rain now performs its own colour grading to match the scene's
colour grading (unfortunately I don't see another way of matching the
colours)
The limitation of the above is
that the values of BRIGHT_PASS_THRESHOLD and BRIGHT_PASS_OFFSET
(HDRLighting.fx) can never change.
The result is a slightly different style of bloom / glare, the kind of
soft-focus look that was used in Fable.
Luckily for me, this suits both the misty environment and the dream
setting.The glares are going to be
designed properly at a later point....
A slight problem with the new
rain lighting is that it turns black at the very edges of the screen.
TOFIX: Noticed that rain moves
with the player / viewpoint
3DVisor: Tried using the
independent left/right RGB gain settings to create a perfectly ghost-free
anaglyph mode.Unfortunately the
ranges of these settings don't allow complete filtering of colour channels.However they should be sufficient to work
as ghost reduction for a player wearing anaglyph glasses under the
3DVisor.I might well give that a try
at some point, crazy as it sounds.
TODO: get rain working in
anaglyph mode<-done(6th)
Passed the 2000 hour mark.I don't know if that's good or bad.
4-Jan-2008
2.83333
Quadro: 14.0
didn't have the missing patchwork decals, and really didn't seem to have the
over-keen bright pass either.Tried
fixed & floating-point lighting; both worked perfectly.Rain seemed to go slightly odd-coloured
when I brought the window size up to 800*600.
TOFIX?: Are rain impact splashes
double-bright on 8800??Would that
make sense, with the brightness bodge?
I'm going to ignore the missing
patchwork bug till I can get hold of an 8800/Vista to test on.No idea what's going on there.
Control improvements (eg. for VR
mode) will be left till version 15 I think.
Had a think about how to
construct the menu tree.Decided to
make each blob a triangle fan, so that joining the branches to them will be
fairly easy.The glow/feather round
the edges will all be done with textures.I'll create the impression of 3D shape using normal maps.
I decided I didn't want to go down the route of rendering the tree as an
offscreen mask and blurring it to get the feathered edges - too expensive and
any lumpiness in the edges due to limited res would ruin it.
Considered building it as a fully-3D model in Max.I'm fairly sure I'd be able to do this, with
a bit of experimentation, but decided not to for a few reasons.The 3D shape doesn't need to be deadly
accurate anyway (as proven by the fact that the mockups looked fairly
convincing).
MENUS: tree shape squidging
should all be done in the shaders (with the blob verts themselves kept in
vertex buffers that change as little as possible).I suppose it'll need to be in the VS so
that branches stay connected.
Added MenuTree.cpp & .h
containing the menu tree class and (temporarily) the classes for the menus
and menu items.
I'm using the prefix CMI for menu item classes, eg.
CMICheckbox.I expect I should really
be using namespaces or something instead....
Updated the coding conventions doc.
CMIRenderTarget: TODO: (ideas at least)
- works like a spinner but has two drag zones, one for each dimension
- RMB changes the units of the dimension under the cursor (see
mockup)
- label flashes / pulses when the control is edited; clicking on the label
applies the change and puts the label back to normal.
- maybe menu tree fades out while the values are being edited (to give a
clearer view of how things are looking) ?
- for the benefit of total-VR mode, spinners need to be mouse wheel
compatible (and there should NEVER be any need to use the keyboard - although
the ability to type values in would be a nice bonus if it's easy to do)
3-Jan-2008
0
Got M's recommendations for a
new PC.
2-Jan-2008
1.75
A couple of things I noticed in
v14 on 8800 (on Vista) :
1. some ground patchwork decals appear to be sorted wrongly.(BC car park & pavement, east manhole
cover, John St. north yellow lines & kerbstones are all missing)
2. The brightness bodge that I added for nVidia cards works as intended,
but the bright pass contains more stuff on 8800 than it does on mine (which
makes sense, because the adjustment is in the 'finalMultiplier' which is done
after everything else).
The most important thing for me
was that v14's VR mode worked fine on Vista, with stereo and headtracking.
Tested on a 16:10 monitor
(setting aspect ratio in the profiles accordingly), all worked fine.
Can't think what the reason
would be for problem 1 above.Tried
forcing zwrite & ztest off when drawing all meshes, they still came out
fine on mine.Tried removing the qsorts
and they still appeared ok (because there isn't much overlapping yet).
Tried reversing the draw order (first in the draw loop, then in
CMeshFrame::QSortCallback_VerticalDepth), to check that this gave the result
I expected.
Maybe it's not a problem with the sorting but with the culling.TODO?: Maybe test a v14 exe with no
patchwork culling?
TODO: option to use the current
desktop res in fullscreen
20-Dec-2007
5.66667
CBlockParser::ReadEnumeration
now correctly ignores trailing whitespace in the property value string.
Flickering, glitchy lighting
actually looks really cool, especially with the motion blur.
TODO?: Just before the enemies arrive, all the streetlights should go crazy
(flickering and even flickering to different colours - like the enemies
disrupt the mains somehow.That way,
you can tell when enemies are coming by the lights glitching, giving you an
incentive not to shoot them all out.Nice!Using the lighting as a
gameplay device is definitely a good idea.Or do something with lightning.Maybe that's their attack?!)
After a bit of history-hopping,
I found the source of the crazy lighting glitch I was getting.I'd added a comment at the end of the
"RTForceFloat: false"
line in default.txt, and CBlockParser was ignoring lines with '//' anywhere on the line!
CBlockParser::ParseBlock now
correctly handles trailing comments.
#revision 825
Options: Testing render-target
property values with whitespace:
"frameRT:[/1<tab>,<tab>/4]"
"frameRT:[/1,/4<tab>]"
NOTE: HDR_InitApp has got nothing to do with HDR; it's legacy GUI setup from
the HDRLighting sample.
TODO: remove it!!!<-DONE (8th Feb 08)
Oh wow,
CNoctambuleApp::RestoreDeviceObjects already calls COptions::ApplyAll.I'd forgotten about that.
So it is safe.
CHeadTracker::Init now calls
COptions::ApplyAll after
choosing a headtracker device and loading its sub-profile.
This is much nicer than applying a sub-set of the options.
#revision 826
COptions::ApplyBeforeDeviceCreation
now updates DXUT according to the options:
calls DXUTStartFullScreen and DXUTSetFullScreenRes.
TEMP: Right mouse button now
toggles eyes (even in master) when in frame-sequential stereo mode
TOFIX: disabling HDR\curve in
Default_HeadTracking_3DVisor.txt went badly wrong....
Checked that it works ok (in 3DVisor mode) when disabled in default.txt
#revision 827
Going for version 14.0 ..... (please work!)
TODO: use a release EMADevice_DLL.dll !
Tweaked
colourGradingMap_default.png by blending on 33% of
colourGradingMap_3DVisor.png on top.Haven't tested it yet ;P
Disabled HDR\curve in
Example_speedyBare.txt
#version
14.0
Released version 14.0
19-Dec-2007
3.66667
TODO: very low priority, but
eventually, I should handle the headset being connected & disconnected
during play.
3DVisor: I've got 4 USB ports on
the back of the PC.The headset works
fine on the bottom two.On the top
two, it gets power (displays work fine), but says "new hardware detected,
Z800, blah blah" and fails to connect.Why is that?
Also, TrackIR does something very similar in the leftmost of the two USB
ports on the side of the case.
I think this was happening
(genuine 3DVisor headtrack connect error), but it's not happening
now...
unconnected
can't connect to headtracker -> CANCEL
can't find headset
plug in
CANCEL
headset found.try headtracker again?
YES
can't connect to headtracker
retry several times, then cancel.Game starts with no headtracking.
I'm leaving any polishing of the
headtracker connection stuff till after the build.It's plenty sturdy enough now and I don't
have time to get sidetracked.
Actually I'm just going to leave it the way it is.It's a matter of opinion which way it
should work, and the current way avoids unnecessary create-then-delete
nonsense.
Kewl, CHeadTracker::Init now has
a Thunderdome-style while loop for choosing between headtracker
hardware!
// PP(19.12.7): up to two devices enter; up to one
device leaves
Testing CHeadTracker::Init with
no hardware hint (default)
- no devices: searched for 3DVisor then searched for TrackIR
- TrackIR: searched for 3DVisor then found and used TrackIR
- 3DVisor: found
and used 3DVisor (didn't search for TrackIR)
- both: found and used 3DVisor (didn't search for TrackIR)
Testing CHeadTracker::Init with 3DVisor hardware hint
- no devices: searched for 3DVisor then searched for TrackIR
- TrackIR: searched for 3DVisor then found and used TrackIR
- 3DVisor: found
and used 3DVisor (didn't search for TrackIR)
- both: found and used 3DVisor (didn't search for TrackIR)
Testing CHeadTracker::Init with OptiTrack hardware hint
- no devices: searched for TrackIR then searched for 3DVisor
- TrackIR: found and used TrackIR (didn't search for 3DVisor)
- 3DVisor: searched
for TrackIR then found and used 3DVisor
- both: found and used TrackIR (didn't search for 3DVisor)
(above) Checked
that all error messages were present and correct.
(below) assertion failure when
starting-up in TrackIR mode with the trackclip out of view.Added a milestone bodge (bailout) and a
REMIND message for now.
Now copes fine with the trackclip starting out of view.
When the game selects a
headtracking device, it now loads and applies one of two options profiles
containing just the default headtracking settings for that device:
Default_HeadTracking_3DVisor.txt
Default_HeadTracking_OptiTrack.txt
These profiles can be edited by the player, but
TODO: make them overrideable by custom.txt, otherwise the player could lose
the default data.<-FIXED(below)
#revision 822
...It now also loads Custom_HeadTracking_OptiTrack.txt /
Custom_HeadTracking_3DVisor.txt straight afterwards, allowing the player to customise without
losing any data.
The default profiles are now kept read-only and the custom ones are
persistent.
Default_HeadTracking_3DVisor.txt now sets up all the 3DVisor default settings, not just headtracking.
COptions::ApplyBeforeDeviceCreation is now called afterwards (instead of just ApplyHeadTracking).
This is precarious and needs to be improved after v14.<-FIXED(20th)
TODO: should also rename these profiles to prevent confusion.
TODO: Fix current crazy lighting
glitch.<-FIXED(20th)
18-Dec-2007
3.5
TODO for v14:
- fix speed-halving in frame-sequential mode<-Fixed(19th: CNoctambuleApp::GetFrameTime. todo: improve?)
- keyboard turn mode, mouse turn option
- mouse-only controls ("total VR" mode)btw, TotalVR.com is taken
- 3DVisor yaw reset thing<-done
- what's going on with that post-build step in
master config?<-TODO: use a release
EMADevice_DLL.dll !
Managed to terminate the 3DVisor
headtracker thread, but only by calling TerminateThread, which is dangerous and not recommended (I think it's quite
safe for my purposes though).
Tried using a volatile 'exit' flag but found that the headtracker thread
doesn't get polled during the shutdown stage.It can be exited smoothly in this way while the game is running, but
not while it's shutting down.Why is
that?
#revision 819
Headtracker thread exit codes: 2
= in response to the exit flag, 3 = forced by TerminateThread
Checked the behaviour of
'assert' in a master build.It's
better than I'd expected, and perfectly safe:
-does work
-does evaluate the expression (eg. function call)
-error box displays source filename, line number and expression
-error box even appears and is functional in full-screen!!
-error box options are abort (quits smoothly), retry (lets you debug the exe, although I think this option only
appears if you have certain stuff installed), ignore (closes the box and carries on).
The V macro, in a master build,
evaluates its expression but ignores any failure.
TODO: switch to using my own
assert macro (give it a unique name, not ASSERT).
C3DVisor::InitHeadset now
"enables the external monitor".
I'm not yet clear on what the benefit of not "enabling the external monitor" is.But I don't want my monitor to go dark when
I'm using the headset (which is the only effect the option has, as far as I
can see).
I've always left the monitor enabled using the utility app, but once or
twice I think I've seen it lose/ignore the setting somehow.
3DVisor: unwanted processes are
now killed before every
connection attempt (see C3DVisor::CreateDevice)
3DVisor: Managed to get an
infinite loop inside EMA device construction, by disconnecting the USB a
second or so after launching the program.
Not much I can do about that, and it's not a worry anyway.
#revision 820
3DVisor init is now almost right, but there's a slight
loophole.I'm about to fix this
now.<-nah, after the build
(TODO: need to create a device without headtracking first, to establish
whether or not the unit is connected....Messy, but I don't see another way of presenting the right information
to the player in the right order.)<-nah, see
19th
17-Dec-2007
3.66667
String table: I'm using the term
"image" to mean render target in the English strings.
Fixed another stringtable
parsing bug that caused some strings to go missing.
TODO: default profile should set
up the headtracking values for Z800, not for OptiTrack ?
Need to decide what the game does about choosing between headtrackers (each
with their different default values).<-ALL
SORTED(19th)
TODO: OptiTrack headtracking on
a separate thread (use headTracking\separateThread).
TODO: terminate the headtracking
threads properly.<-done(18th), but see note
Added some OptiTrack dialogue
boxes.
3DVisor: If the connection
fails, the game now does one automatic silent retry before it starts
displaying error messages.
16-Dec-2007
10.25
(below) Spent a couple of hours
mocking-up the pause menu in Photoshop.It was only afterwards that it reminded me of neurons, and of bubbles
seen underwater, and of a heart.
Anyway, the menus are going to take the form of these
translucent blobs, connected together in a sort of a tree around the player's
head.It's a sort of sci-fi
"organic computer display".
- Clicking on a 'link' makes the tree rotate so that the target menu is
centred in the field of view.
- Another option, when in VR mode, would be that the tree always stays put
and that menus just change colour as they become current (when a 'link'
directs to the menu or a control on the menu is used).You'd be able to see the colour change
along the branch from the menu you're currently looking at.
...
...
-For mega brownie points,
the tree would contain a Navier-Stokes fluid
simulation, so coloured dye could run through
it.The dye could be attracted to low
pressure in the current menu or something.The dye could have a different refractive index from the clear
fluid.I'm not likely to do this dye
thing though - it'd be tons of work and there's better ways I could be
spending my time.
-The tree should write noisy values to the velocity
buffer, to cause slightly jumbled motion
blur.This will add a layer of
dreaminess straight away.See
flashback sequences from Underworld.
-The text boxes might also want their own glitching
effect.This will provide some movement as well as making the text look
'dreamy'.Letters should sometimes
change into different letters of similar width.Be careful to avoid hackneyed text effects,
eg. the skateboard-game jitter (radical!)
-The tree must gently squidge around / pump or something; must not be frozen.The current menu could beat like a
heart.
-Tree rotation needs to be motion-blurred.
I think I've got enough to be
going on with; I'll play the rest by ear.
"
Global.h: Disabled warning
4345:
"behavior change: an object of POD type
constructed with an initializer of the form () will be
default-initialized"
Putting the 3DVisor headtracking
on a separate thread has brought my framerate from around 20 to around
33.
(800*600, debug, default settings).
Based on those numbers, it must take about a 50th of a second to get a
reading from the headtracker, regardless of the PC.I should time this properly, but I've got
more interesting things to do.The
important thing is that it's not slowing the game down anymore.
With nearly all the effects
stripped out, I'm managing to keep the VR mode above 60 (most of the time) in
Master config now, which lets me at least get some idea how the game should
feel (and it's ver' nice).Need to upgrade this PC to get the full
experience.
Fixed the head-tracked stereo
separation for 3DVisor.It wasn't
properly accounting for the display being head-mounted.
TODO: for TrackIR, use the old method (from SVN) and test that it's
correct.Check display\headMounted.
15-Dec-2007
10.5
Found some occasional flickering
artefacts in the floating-point lighting (800*600) :(
Floating-point lighting on this card also removes various filtering
(including some full-screen filtering that shouldn't be
there; TODO: investigate<-FIXED(12.4.8))
Started setting up the string
table functionality.
TODO: For the likes of the credits, the stringtable
parser should be able to invent the absent identifiers by appending the UK
string (suitably processed) onto the last identifier it found.
eg. "ESTRING_CREDITS_THANKS_CelineAccabat"<-DONE
#revision 806
Set up a Tools\StringTableParser
project and added it to SVN
SVN: Never source-control .NCB
files!
Stringtable:
0x0D, 0x0A = end of row
0x0A = newline
0x09 = tab (cell delimiter)
I'm opening stringtable.txt in binary mode because in text mode the 0x0D characters get filtered out
#revision 808
String table parser is now
working nicely.
Added function UtilGetWideString
(now used by UtilGetTempWideString).
TODO: CStringTable::StringW ?
Added function UtilMessageBox
(takes in stringtable indices and control flags; returns response code).Very handy.
Teehee!In the French 3DVisor user guide, someone's
accidentally left a little green text box that says "Testmuster" !
Set up a nice little network of
dialogue boxes to handle any & all problems during initialisation of the
3DVisor.Spent a wee bit of time
translating these into French.
3DVisor: current firmware
version is 6.3E
3DVisor:The new utility, Lab Tool, seems to lose /
give up its own connection when the game starts, rather than preventing the
game's connection.Nevertheless, I've
added it (LabTool.exe) to list of processes killed before the headset
connection is attempted.
#revision 810
Pleased with the stringtable
system; it's real smooth.Dialogue
boxes appear in the right language as specified by the 'general \ language'
option (which defaults to French bien sûr).
13-Dec-2007
3
3DVisor: considered making the
game change the headset's timeout values, but decided this is something that
the user won't want me to mess with.So I'll continue using the 'keepalive' signals (these won't slow the game,
so they can't interfere with frame sequential stereo).Even resetting the EEPROM doesn't change
the timeout values set by the user.
3DVisor: when the you first
connect to the headtracker, its current yaw is is set as the zero yaw by the
eMagin system.
3DVisor: [update: almost] Never fails to connect to
headtracker when it's in standby (flashing green) mode when you try to
connect.
Wanted to try something that might guarantee a successful headtracker
connection:
- connect firstly without headtracking
- call GoSleep
- clean up
- connect again with headtracking.
See SDK source for details of why this isn't possible :(
I'm a bit paranoid about
mentioning any eMagin SDK stuff in this log; the SDK is all marked
confidential, and I only managed to get hold of it after I'd contacted eMagin
from my work email account.
Y'know, the headtracker
connection used to work every time anyway.Is it only since installing Lab Tool that it's started failing?I have a feeling that did install a new
drivery type thing.
Another change in behaviour (I think) is that after an unsuccessful
connection, simply putting the headset into standby makes it work next time
([update: nearly] always
works).Used to have to disconnect the
USB lead & re-connect it (unless I just forgot to try the standby button
before ?).
3DVisor yaw now resets when
exiting pause mode.TODO: neaten this
(should orient the player to the current virtual head yaw; currently snaps
the virtual head yaw to the current player orientation).
See CHeadTracker::ResetTransform
TODO?: Might sound silly, but
I'm quite tempted to design a jetpack into the game, and maybe flying
enemies.Flying around even with the
flycam, in VR mode, is amazing.It's
partly because the headtracking gives you a real sense of height when you
look down (and I'm not even in stereo mode yet).
Headtracking is great.It connects
another one of your senses into the game: the sense of balance /
direction!And the more senses you
connect to the game, the more it feels real and engages your instincts.
Complications with a jetpack would be:
- requires more advanced building collision (probably need this for weapons
anyway?)
- occlusion culling gets more complicated (and doesn't help much once
you're above rooftop level).
12-Dec-2007
3.7
Fixed the stuttery
headtracking.I'd left an "if(
(g_game.frameNumber&4)==0 )" in the headtracker update, duh!
That's a relief.
(below) Made a new &
improved grading map for 3DVisor mode.Although obviously it's not clear from a screenshot, this comes out
nice & vivid while preventing the over-contrast that these little
displays would give if using the default grading.
(left to right: no grading, grading for 3DVisor, grading for CRT)The CRT one is a bit harsh at the moment -
you lose too much of that subtle mist.I'll bring it back a bit.
I'm fed up with the bug where
everything's double-bright on nVidia cards; it gives a bad first impression. I'm guessing it simply
happens when using float RTs for the lighting.<-no,
see below.
TODO: test floating-point vs. forced fixed-point on an nVidia card.
TODO: bodge: halve the HDR\finalMultiplier on
all nVidia cards<-DONE(13th, in
COptions::ApplyAfterDeviceCreation)
CBlockParser::ReadBool now
accepts '1' as true.
WEIRD: I tried setting lighting\RTForceFloat to true on my
x1900, which I was pretty sure couldn't blend to floating-point RTs.Erm, but it seems to be working
perfectly.And it's definitely using
floats; the banding is practically invisible, same as on 8800.
Maybe when I tested this last time, I was trying to blend to
D3DFMT_A32B32G32R32F (which comes out black).With D3DFMT_A16B16G16R16F, it seems 100% fine.
Maybe I tested ALPHA blending last time, rather than additive?(the lighting uses additive blending to
build-up its offscreen images).
Here's what I'll do:
lighting\RTForceFloat now defaults true.
TODO: test 3DVisor head-tracking
on a separate thread
TODO: need an eye-swap calibration mode for 3DVisor (display 'left' &
'right' over the relevant eye)
11-Dec-2007
2.66667
"VxDs are not usable in
Windows NT or its descendants. Starting with Windows 2000, these operating
systems use the Windows Driver Model (WDM)..."
Summarised my quest for
framerate-independent stereo, on the eMagin forums:
http://www.3dvisor.com/forum/viewtopic.php?t=1036&start=0&postdays=0&postorder=asc&highlight=sync
If only I could find that WINx3D SDK !!!
Right then, this time I'm really giving up with it, until
such a time as fresh information comes in through the forum.
Now turning my attention to Version 14: Xmas
Edition™ instead!
#revision 801
SVN: Finally managed to untangle
SVN and add the driver projects for safekeeping.
Noticed that 3DVisor.exe, even
just sitting in the system tray, is continually spitting-out a debug line
such as:
[4076] VendorID: 9ac8, ProductID: 6b
I wonder what it means.
(searched wdk help for wdm.h)
ms-help://MS.WDK.v10.6000/Kernel_d/hh/Kernel_d/WDMIntro_d5b4fea2-e96b-4880-b610-92e6d96f32be.xml.htm
"Kernel-mode drivers that follow WDM rules are called WDM drivers. All
WDM drivers must:
Include wdm.h, not ntddk.h. (Note that wdm.h is a subset of ntddk.h.)
..."
I need to include wdm.h in order for video.h to compile.
The avssamp sample
driver is WDM one, and I can build it, but when I try to load it using
InstallDriver.exe it fails with ERROR_SERVICE_DISABLED.
If I do manage to convert DemoDriver to WDM,
will I get the same load error??<-No, converted
it now and it still loads
"The I2C bus has only two
wires: the serial clock line and the serial data line.
Reading and writing data bits to the I2C lines on the display adapter is hardware dependent, so the vendor-supplied video miniport
driver must provide the functions that instruct
the display adapter to read and write the individual bits."
Hmm yeah, I've been a bit
thick.WriteDataLine etc aren't
functions I can call,
they're functions I can implement.
If I'd paid closer attention to the docs, I
would have seen that they're callbacks provided by the driver (that's me).
"VideoPortDDCMonitorHelper
implements the details of reading the EDID structure according to the I2C
specificaiton, but must call back into the video
miniport driver to read and write individual data bits to the I2C serial clock and data lines.
The four functions, implemented by the video
miniport driver, that read and write individual
bits to the I2C clock and data lines are ReadClockLine, ReadDataLine,
WriteClockLine, and WriteDataLine. When the video miniport driver calls
VideoPortDDCMonitorHelper, it supplies pointers to those four functions in
DDCControl->I2CCallbacks."
Went back and had a look at
DrYak's info about syncing the stereo:
http://www.3dvisor.com/forum/viewtopic.php?t=1036&start=0&postdays=0&postorder=asc&highlight=sync
Downloaded eMagin's Lab Tool
thing.Yak's idea
about using different sync values between the two eyes sounds very
interesting...<-it isn't possible to use
offsets that big
Allowed the Lab Tool to remove
3DVisor Software Utility, as it seemed to want to do.<-re-installed
it
Lab Tool reports that my headset
has the "3D Commands" capability.Not sure what it means by that, probably just means the enable/disable
3D mode command.
3DVisor: phase at around 18
seems to work well for the stripes calibration test.
clock=2047/4095 = show left side
of screen
Tried fiddling with phase, clock
and position settings, but they don't allow a big enough offset to do a
split-screen-style stereo mode.
TODO: find out about SciTech
GLDirect.Is it a forerunner to Snap?
TODO: maybe get hold of that
sync-doubler VGA pass-through thing that comes with those LCD shutter glasses
at work.I think that will actually
alternate the Z800 (refreshing at 120Hz).But I don't think it helps me in trying to send my own image
pair.Would be an interesting
experiment anyway (in fact, wouldn't this make every game work with the
Z800?)
9-Dec-2007
10
Hmm, I lost rather a lot of time
to those two gigs (plus hectic work stuff).
Need to keep an eye on how much time gets lost practising next year.It might become necessary to pull-out of
band activities 'early'.
I'm probably going to give up
trying to do the framerate-independent frame-sequential stereoscopy thing; it would be really nice
but there's more important things to take care of, and time is getting
short.
So if you have a frame-sequential headset, and you want to run the game
stereoscopically, you'll just have to make sure your PC can keep the game
above 60fps.<-currently trying to get it
working using the WDK
The annoying thing is that mine can't :(
I should look into upgrading....
Of course, it would be a very
fair criticism to say that the game simply runs too slow - that's the truth
of it.
But, although it sounds flippant, my time is far better spent making the
game look pretty than making it fast, because the PC/GPU setups people use to
play the latest games are fairly monstrous anyway and they can handle
it.No point spending a load of time
optimising when the target audience was already able to run it at 60 (which
they are) - it's invisible work.
below: here's the VBE doc for
the I²C functions that should let me control the DDC
pin:<-can't generate interrupts from Win32,
see below
Function 15h, Sub-function 11h -
Begin SCL/SDA control
Function 15h, Sub-function 10h - Report VBE/SCI Capabilities
Function 15h, Sub-function 14h - Write SDA data line
[Function 15h, Sub-function 13h - Write SCL clock line]
Function 15h, Sub-function 12h - End SCL/SDA control
Example French function
descriptions:
"lit les données EDID du moniteur principal"
"ouvre le fichier DUMP"
"écrit le buffer des données EDID dans le fichier DUMP"
TODO: have a dig around on some French coding sites, get more of a feel for
commenting & naming
"Generating interrupts only
works when running a 16 bit program... Generating an interrupt from a Win32
program
will cause an access violation."
"You cant make BIOS calls in windows from a user mode app...
You can use a published API.......lookup DeviceIoControl on
MSDN...."
"at http://www.geocities.com/SiliconValley/Haven/4824/vxdprg.html Bill
Alexander has free source code for a VxD to do BIOS calls. Looks like I'll
have to recompile it with the Windows 95 DDK though (it's Windows
98)."
"Win32-based applications cannot use MS-DOS and BIOS interrupts"
...So it's not possible for me
to call BIOS functions directly (which would make things very, very
easy)
Using 'INT' causes an exception like it says.
The only route left to try now
seems to be the I²C functions in the WDK.Can't remember how far I got with them before...
"WriteDataLine sets the I2C serial data line to high or
low."
(see 28th Oct 07)
Needs to be a separate driver I think
Writing a device driver for
Windows :
http://www.adp-gmbh.ch/win/misc/writing_devicedriver.html
Essentials Of Building Windows
Drivers :
http://www.wd-3.com/archive/WinBuild.htm
I've now managed to get the
example driver DemoDriver (from the Essentials article above) to build, and
load, and be successfully called from the game.Had to update it here & there because
it was written for the 95/98 DDK.
I just haven't found a good way of debugging
the driver code yet.<-Now using DebugView, see
below.
It's definitely running the driver code, because if I add an ASSERT, in
'checked' configuration, it causes my PC to instantly restart.
Installed a thing called
DebugView
(http://www.microsoft.com/technet/sysinternals/utilities/debugview.mspx),
which catches debug output, like Watson does.
But it doesn't catch any of the DbgPrint calls in the
driver code.<-Yes it does, just need to tick
'Capture Kernel'.
DemoDriver is a kernel-mode
driver.
Just so I don't forget:
Firstly run one of the WDK build environments for XP:
C:\Documents and Settings\user\Application Data\Microsoft\Internet
Explorer\Quick Launch\WDK 6000\Build Environments\Windows XP
To build the driver:
cd C:\WinDDK\6000\src\WinBuild
build -cef
To load the driver:
cd C:\WinDDK\6000\src\WinBuild\libchk_wxp_x86\i386
InstallDriver DemoDriver load
To unload the driver (must do this then re-load
each time it changes) :
InstallDriver DemoDriver unload
Added DebugView to the Visual
Studio 'tools' menu.
Tried to add all the WinBuild
(example driver) stuff to SVN, but couldn't quite remember how to do
it.
I'll just wait till I've set up the new driver under the Noctambule folder,
that'll be easier.
Tomorrow, if all goes well, I
should be able to get that DDC pin doing its thing.
If the headset responds, that's obviously wonderful.If it doesn't, I might try and get hold of
a multimeter to check that the signal is coming through.
15-Nov-2007
0
(below) possible enemy
inspiration: really gruesome baby dolls, suitably nightmarish:
I'm completely tied-up with the
band at the moment, so I'm not going to be able to make much/any progress
till December.
I will almost certainly do a build in time for Christmas with the following
features:
- native z800 head-tracking support (if you haven't seen this headtracker natively supported yet,
you'll be amazed how good it is).
- z800/x800 (frame sequential) stereoscopy mode - Note: this will only work if your PC can keep the game above
60Hz.
C:\scitech\mgl60r23\bin-win32>dmake
C:\scitech\mgl60r23\examples\snap\graphics\hello\makefile
'C:\scitech\mgl60r23\examples\snap\graphics\hello\makefile'
is up to date
?
C:\scitech\mgl60r23>my_start-sdk.bat
Release build enabled.
Open Watcom C/C++ 1.0 Win32 GUI compilation environment set up.
C:\scitech\mgl60r23\examples\snap\graphics>
C:\scitech\mgl60r23\src>dmake
build
=======================================================================
** BUILDING: SciTech SNAP SDK libraries for current compiler
=======================================================================
START: Building PM Library
wcc386 win32\pm.c
dmake.exe:Error -- wcc386: No such
file or directory
dmake.EXE:Error code 129, while
making 'targets\pm'
dmake.exe:Error code 255, while
making 'single'
C:\scitech\mgl60r23>my_start-sdk.bat
Release build enabled.
Visual C++.NET 7.0 32-bit
Windows compilation environment set up
C:\scitech\mgl60r23\examples\snap\graphics>
C:\scitech\mgl60r23\src>dmake
build
=================================================================
** BUILDING: SciTech SNAP SDK libraries for current compiler
=================================================================
START: Building PM Library
....
k_cp n_ga.lib C:\scitech\mgl60r23\lib\release\WIN32\vc7
================================================================
DONE: Single compiler SNAP SDK build completed
successfully!
mkdebug and
mkrelease (replacing dmake build) are also both
successful
French strings for a doom-type
game:
C:\scitech\mgl60r23\examples\mgldoom\d_french.h
Couldn't get the Doom example to
build:
C:\scitech\mgl60r23\examples\mgldoom>dmake
...
LINK : fatal error LNK1181: cannot open input
file 'mgl.lib'
dmake.exe:Error code 130, while
making 'doom.exe'
There is no mgl.lib anywhere :/ (?)
C:\scitech\mgl60r23\examples\snap\graphics\gatest>dmake
...
LINK : fatal error LNK1104: cannot open file
'gatest.def'
LINK : fatal error LNK1141: failure during build of exports file
dmake.exe:Error code 130, while
making 'gatest.exe'
There is no gatest.def
Made a gatest.def that just
contains
NAMEGATEST
C:\scitech\mgl60r23\examples\snap\graphics\gatest>dmake
cl /nologo @C:\DOCUME~1\user\LOCALS~1\Temp\mk000001 Creating library gatest.lib and
object gatest.exp
LIBCMT.lib(wincrt0.obj) : error LNK2019: unresolved external symbol _WinMain@16
referenced in function _WinMainCRTStartup
gatest.exe : fatal error LNK1120: 1 unresolved externals
dmake.exe:Error code 130, while
making 'gatest.exe'<-needed to change
mystartsdk.bat to use 'c32' instead of 'w32', see below
_WinMainCRTStartup defined in
build\intel\mt_obj\wincrt0.obj
dmake -u to force rebuild
mainCRTStartup (or
wmainCRTStartup) ---- An application using /SUBSYSTEM:CONSOLE; calls main (or
wmain)
These example apps are CONSOLE
apps; they have a 'main' rather than a 'WinMain'.
So the compiler (cl) arguments want to contain /link /SUBSYSTEM:CONSOLE, as
shown here:
C:\scitech\mgl60r23\examples\snap\ddc\ddctest>cl ddctest.c /link /SUBSYSTEM:CONSOLE
cl is being called from dmake.exe<-needed to change
mystartsdk.bat to use 'c32' instead of 'w32', see below
C:\scitech\mgl60r23>mystartsdk.bat
Checked debug build enabled.
Visual C++.NET 7.0 32-bit Windows console compilation environment set up
C:\scitech\mgl60r23\examples\snap\graphics>
Built the SDK libs with w32
again.
Well, I've now got as far as
building an exe (snap/ddc/ddctest.exe) and actually being able to run
it.
I now get this error message:
---------------------------
Fatal Error!
---------------------------
Unable to load SNAP device driver!
---------------------------
OK
---------------------------
COOL!I can attach the debugger to the
ddctest.exe process and get a call stack.
ddclib.c \ LoadDriver
looking for ddc.bpd
Fixed that (set up a
C:\windows\system32\snap folder), but now I get this:
---------------------------
Fatal Error!
---------------------------
Invalid license!
---------------------------
OK
---------------------------
Up till now, I thought this thing was free :/<-it is free (GPL) see license.faq and license.gpl
TODO: get C:\scitech into SVN
C:\scitech\mgl60r23>mystartsdk
Checked debug build
enabled.
Visual C++.NET 7.0 32-bit Windows compilation environment set up
C:\scitech\mgl60r23>SET CHECKED=0
C:\scitech\mgl60r23>SET DBG=0
C:\scitech\mgl60r23>mystartsdk
Release build
enabled.
Visual C++.NET 7.0 32-bit Windows compilation environment set up
Hmm, I can't seem to build the
SDK libs anymore, fails near the end.It's fine until this point:
START: Building Techniques Class Library
cl /nologo @C:\DOCUME~1\user\LOCALS~1\Temp\mk000001 /c dlist.cpp
dlist.cpp
cl /nologo @C:\DOCUME~1\user\LOCALS~1\Temp\mk000100 /c list.cpp
list.cpp
cl /nologo @C:\DOCUME~1\user\LOCALS~1\Temp\mk00022a /c hashtab.cpp
hashtab.cpp
C:\scitech\mgl60r23\include\tcl\hashtab.hpp(46) : fatal error C1083: Cannot
open include file: 'iostream.h': No such
file or directory
dmake.exe:Error code 130, while
making 'hashtab.obj'
dmake.EXE:Error code 129, while
making 'targets\tech'
dmake.exe:Error code 255, while
making 'single'
I don't seem to get an accurate
callstack when I get the 'invalid license' message.Can't see where the message comes
from.Can't find that string in any of
the files.
isvLicense variable is NULL though
license.h just defines EMPTY_LICENSE, whatever that means
Downloaded the pre-build MGL
Doom demo; tried to run it but it instantly restarted the PC.
C:\WINDOWS\system32\snap\config\snap\config\ddc.log:
"Unable to validate license!"
emailed license@scitechsoft.com
asking about the licence problem.I
don't even know if the company still exists, and even if they do I doubt
they'd answer my email, but you never know.
C:\correspondence\scitech
Hmm, when I add a call to
GA_getDaysLeft into ddctest.c, before the call to LoadDriver, I get
this:
---------------------------
Fatal Error!
---------------------------
Unable to load nga_w32.dll!
---------------------------
OK
---------------------------
Hmmmmm........
3-Nov-2007
7
Installed Ad-Aware 2007
Free
Installed Spybot S&D
#restore
In Computer Management:
Disabled my printer port (in the Computer Management program), to see if it
stops the 'found new hardware' thing every time I start Windows.<-didn't help; I've re-enabled it.I'm too scared to 'uninstall' it because I
don't know how I'd re-install it.
...also disabled the Windows
Indexing Service.
Why are lsass.exe &
csrss.exe constantly chipping-away at my hard drive?I gather this is normal, but it's a bit
irritating.
Installed Win3D Drivers (which I
think use redraw-synced page flipping), but couldn't get the test app to work
at all.And they didn't include an SDK
or any source.
Even in anaglyph mode it gives the same message: "this image cannot be
queued for this eye".
3DVisor: Here's a hardware mod
that lets the visor take in dual monitor inputs (but it seems to require the
removal of the headtracker hardware).Presumably gives you 60Hz 3D.
Not useful for me, just interesting:
Been raking around on t'Internet
trying to find the mythical WINx3D SDK.Looks like exactly what I need, but no-one seems to know where it can
be found anymore.
"...for page-flipped stereoscopy, the start address must change every
time a vertical sync occurs on the video card. This is the primary purpose of
WINx3D - to keep the image that is displayed on the video card synchronized
with the VR gear. The DirectDraw Flip() function is inadequate for this"
Sent an email to a couple of people who were also looking for it, in case
they still have the same email addresses and managed to find it:
So it was an evening of
detective work.But I can justify
spending a bit of time on this - if I can manage to put all the pieces in
place, it'll be amazing.
2-Nov-2007
0
...and it's just as well I did
back up the project!
Just after writing that last entry, I noticed I'd picked up a virus,
started heavy-handedly trying to weed it out, and ended up knackering my
registry - leaving me unable to log in to Windows.
Had to crash on the Floor of Ubuntu while I worked out how to repair the
registry files, but thankfully the procedure is simple and
well-documented.
The registry rollback not only let me back into Windows, but also disabled
the virus, which is nice.Should maybe install some kind of anti-virus now though....<-DONE
Windows: Found that the
system-tray programs to be launched on startup are specified in this registry
folder:
Software\Microsoft\Windows\CurrentVersion\Run
WDK: Getting
Started with Windows Drivers:
http://msdn2.microsoft.com/en-us/library/ms791699.aspx
VideoPortDDCMonitorHelper requires a device extension (PVOIDHwDeviceExtension)
IoCreateDevice supplies the device object &
extension (OUT PDEVICE_OBJECT*DeviceObject)
IoCreateDevice
requires a driver object pointer (PDRIVER_OBJECTDriverObject)
Driver object pointer is received by DriverEntry (PDRIVER_OBJECTDriverObject)
DriverEntry is called when a driver is loaded.
In other words, I don't think I can use VideoPortDDCMonitorHelper within
the game - would have to be in a separate driver.
I wonder if someone has already
written a driver to do what I want to do?.....
This looks like a good
article:
Driver Development Part 6: Introduction to Display
Drivers:
http://www.codeproject.com/system/driverdev6asp.asp?df=100&forumid=262328&exp=0&select=2167158
Downloaded a 30-day trial of
"WinI2C/DDC" (catchy!), which is like an all-singing-all-dancing
version of the driver I need.
No use to me, because it's $450 or something.
Just wanted to use it to test the DDC control of the headset.
26-Oct-2007
2.83333
#revision 770
Tried using D3DPRESENT_DONOTWAIT
to catch dropped frames.Obviously
doesn't do what I hoped it would; IDirect3DSwapChain9::Present never returned
D3DERR_WASSTILLDRAWING.
If QueryPerformanceCounter and
QueryPerformanceFrequency are accurate, then video output is flipping slightly faster than 60 Hz (more
like approx. 60.333 Hz when the game is fullscreen).And the headset is fine with that, because
it just syncs with the monitor signal rather than swapping exactly 60 times
per second.But why is the output not
exactly 60 Hz ?!
And more importantly, how on earth can I sync-up with the actual rate?I've tried all sorts but can't get it to
work reliably.
Several people on the 3DVisor forums have run into the same problem.
Blimey, VGA includes a Display Data Channel / Data clock line (DDC) and the 3DVisor can read this as a signal indicating
which eye to send to!That'll be why
the nVidia stereo drivers didn't give a random eye swap each time - they were
selecting each eye in turn explicitly!And that's how they kept the headset in sync regardless of the
framerate.That's what I want
too.
Some explanation here:
http://www.3dvisor.com/forum/viewtopic.php?t=1036&start=0&postdays=0&postorder=asc&highlight=sync
DDC on Wikipedia:
http://en.wikipedia.org/wiki/Display_Data_Channel
Installed softMCCS, hoping it would let me poke DDC signals out to the
headset.It doesn't seem to be aware
of the headset being there, so it just tells me that my monitor isn't
plug&play and DDC/CI (and all the other stuff) isn't available.I would've expected it to let me send
something out regardless :(
Running softMCCS has made my monitor quite dark now for some reason.
TODO: try using
VideoPortDDCMonitorHelper to write DDC signals (#include C:\WinDDK\6000\inc\ddk\video.h)
25-Oct-2007
1.75
Game now kills AdobeUpdater.exe
on startup.That's the last time that program interferes with my
fullscreen mode!
Still trying to get that
frame-drop handling working....
21-Oct-2007
6.083
TOFIX: (does this always
happen?) when the headset's in standby (flashing green light, screens off),
the game fails to connect to it.Hmm.
Pressing the power button to wake it up first fixes the problem, but it
would be nice if this could be automated.
Noticed that after a headset
init failure, simply re-running the game can make it work properly.Maybe try putting the init in a loop so it
can make two attempts??
#revision 768
Removed option anaglyph\laggingApproximation, to
simplify some code.Anaglyph
stereoscopy is rubbish enough without allowing lag between the eyes.
Website: I was thinking about
getting one or more .fr domains.From
wikipedia:
"There is a French presence requirement and administrative contact
address must be in France; some other restrictions on some of the specific
subdomains"
So the only legal way to get one at this point would be if someone in
France could buy it and act as the admin contact on my behalf (I don't even
think it would be legal for them to transfer the domain to me).So it's not worth the bother.
From AFNIC.fr:
"Le contact administratif est impérativement établi en France et doit
y disposer d'une adresse effective qui lui permette de recevoir des actes
judiciaires ou extrajudiciaires."
Spent ages trying to get the
frame-sequential frame-drop handling working but haven't managed yet.TODO: find out why fps shows as > 60
when running in 1 frame.
18-Oct-2007
3.5
25*25, frameRT=[/2], debug,
windowed, everything switched off (phil.txt):
40fps
+800*600 fullscreen, frameRT still
[/2]
60, +/- 1 occasionally and fleetingly
+C3DVisor::GetHeadsetData:
48 fps
Building the game project:
"Compiling resources...
Linking...
Embedding manifest..." (endless pause)
- same happens after a clean rebuild
- mt.exe maxing the CPU.I cancel
the build
- vsmsvr.exe & devenv.exe each using 50% CPU
- I close both VisualStudios, there's still 2 devenv.exes and vsmsvr.exe,
still doing the same thing.
- I close Excel.Excel.exe still
present in the task manager.
- Fraps app isn't visible, but Fraps.exe process still showing.I try to end its process tree.
- PC goes into super-unresponsive mode (screen updating once every 5
seconds or so; effectively no controls are usable).Needs a restart.
There was no visible harddrive or network activity during any of this.
After the restart:
Opening Noctambule.sln, definitely no other copy of it open:
"IntelliSense information will not be available for VC++ projects
because the IntelliSense database file C:\Noctambule\Source\Noctambule.ncb
could not be opened for writing."
Did another clean rebuilt of
noctambule.sln, and everything's fine now (except no intellisense).3DVisor.exe is running btw, so that's not
the problem.
I do suspect EMADevice_DLL.dll might have been involved in all this though.
#revision 766
+frameRT=[/1], fullscreen 800*600:
60 fps in most places, as low as 47 in the worst case (standing at CRUK
corner looking at BT)
The eye-swapping (caused by
frame drops) makes me feel quite yucky quite quickly.Let's fix that now using a timer...At least I hope it's
the eye-swapping and not just general simulation sickness from the fake
moving and turning while in stereo mode.I haven't set up the stereo convergence properly for Z800 yet either;
that could be relevant.<-(18th: seems fine when
the views are the right way round; I've also set convergence
to 0 in VR mode now)
I still can't believe that thing
I read (might've been eMagin blurb) claiming that the brain copes fine with
flipped stereo views and "makes sense" of the depth.Nonsense!!Complete nonsense.You get the
impression that there's depth information in front of you, but not the sort
that makes any kind of sense.It's
like saying that pushing-in the side of one eyeball slightly turns your TV
programmes into 3D.
Two things that are really
effective in VR mode:
- sheltered hidey-holes -
you get a much stronger feeling of enclosure.A sheltered audio effect should also strengthen this immensely.
- light glares (when
different in each eye, even just occluded in one eye) - when your brain gets
the two non-matching streaky patterns of light it confidently registers that
there's a lightsource there.Will be
great when I add the proper light glares.They'll respond correctly to headtracking too (nice
counter-rotation...)
TODO: fix stereo separation to
take into account head tilt
TODO: for VR mode, a blurry
black screen border would seem to be a sensible idea (to give the impression
that the 'window' is too close to be in focus).This might make the stereo feel more
natural.
TODO: get the frame-drop
handling working!
17-Oct-2007
2.75
(link below) More examples of
that speed-painted art style that I want to use for my next project that I
keep thinking about but will never have time to do.
These are by a bunch of Pixar people who get up at silly o'clock to paint
stuff:
Managed to get frame-sequential
stereoscopic mode working on the Z800!!ie. managed to reach 60fps, or "88" as I like to call
it.
Had to switch pretty much all the effects off :(
Also had to switch off head-tracking, because waiting for the readings to come back takes a
seriously long time.TODO: put that on a separate thread!!!Waiting around for that kind of thing is
crazy!I should really do the same for
TrackIR, if I can be bothered.TrackIR's a complete relic for me now.
Frame-sequential stereo fixes
described in yesterday's entry.
16-Oct-2007
2.5
Why does Visual Studio sometimes
just lose the project from my
solution?
Added option category controls
Added options
controls\mousePitch - Defaults
true.It's false in the Z800 example
profile (no need to pitch with the mouse when you're using Z800, that would
be silly)
controls\mousePitchSpeed
controls\mouseTurnSpeed
TODO: try a cursors-only control
style (side keys to turn while running).This will probably be more suitable for VR mode.Add option controls\mouseTurn.
Oh yeah, last night I dreamt I
was in one of those WWII shooters (presumably called "Men Of Honor"
/ "Honor Of Heroes" / "Band Of Men" / "Medal Of
Valor" or something like that ;)
We were escaping from a sort of POW camp.During the escape there was a bit where I was basically running away
from a guard taking shots at me, knowing that any second one of them would
probably hit.Was quite a tense bit of
action; would be good if my enemies could shoot.Apart from anything it would mean more
sparkly effects flying around the place - always a good thing.
I would also really like
to somehow recreate the fear I got from watching Dr. Who when I was little
(Daleks especially).I reckon a lot of
that fear was created by the audio.
Hmm, my dreams have no sound - speech is present and sometimes there's
(pretty good) music, but no other sounds whatsoever!Maybe it's because I've spent so long
playing games with the sound off (at work), so I'm used to all the silent
action.
Added option category screenEffects.This is for all the screen effects apart
from those belonging to the HDR system.And apart from motion blur, which isn't intrisically a screen effect
the way colour grading is.
Added option:
screenEffects \ colourGradingMap
#revision
763
Removed all the 'topline_'
textures (old WOWvx headers).
There's still copies in C:\noctambule\3dtv
#revision
764
Added folders
Graphics \ Final \ Effects
(eventually I'll move all the effects graphics into here)
Graphics \ Final \ Effects \ ScreenEffects which
now contains the two colour grading maps.
Thanks to all my
forward-planning, it only took a couple of lines of code to get the
frame-sequential stereo mode (mostly) working :)
TODO:
- stop it producing motion blur<-done(17th) now got a "previous final view
matrix" per eye
- stop the strange flickering of the lighting<-done(17th) mist lighting RT wasn't getting cleared between
eyes.
- add an eye-swap key<-(17th: now on right mouse button
temporarily)
- add dropped-frame handling? (so that frames can be dropped without the
eyes swapping)?I think the headset
waits a second or so before switching out of stereo mode (so dropping a frame
here or there shouldn't switch it out).<-(17th: that's exactly right)
15-Oct-2007
0
Er, been really tied-up with
work and the band.
Bought a new aerial photo, from
Bluesky this time rather than Getmapping.
It's tons better!Same res, but much clearer and I reckon
it's also more positionally correct than the blurry, badly-stitched JPG I've
been using till now.This one's a
GeoTIFF, which Photoshop loads just like a TIFF.
This is going to make things a lot easier when I do my next batch of
construction work.
It cost slightly less than the last one, too.I'm really quite glad I found this.Found it through Google Earth.
6-Oct-2007
4.25
Removed the Z800 Software
Utility from my startup programs.<-added it back, to
check that the auto-termination of this process is always working (see
below).
The game connected with the headset on first
attempt this morning, from standby (flashing green) mode.
Re-started the machine, worked first time again, from 2D (amber) mode.
Aha!I just need to kill 3DVisor.exe before trying to connect?
dxstdafx.h: added #include
<tchar.h>
(needs to be included before strsafe.h)
Added UtilEndProcess.
Had to add linker input Psapi.lib for this.
UtilEndProcess reference from
Microsoft docs:
Enumerating all processes:
http://msdn2.microsoft.com/en-us/library/ms682623.aspx
How To Terminate an Application "Cleanly" in Win32:
http://support.microsoft.com/default.aspx?scid=KB;EN-US;Q178893&
#revision 761
On startup, the game now
terminates any and all processes it finds by the name of
"3DVisor.exe"
It now successfully connects with the Z800 headtracker every time.
5-Oct-2007
3.5
Z800: AngX, AngY, AngZ seem to
be angular accelerations.
AccX, AccY, AccZ must be the positional accelerations.I doubt I'll have any use for those, except
maybe for detecting crouch
& uncrouch, not likely though.
Z800: "The headset
includes: 3 gyroscopes, 3 accellerometers and a compass"
Experimented with yaw vs.
compass readings.I can't see any
difference in results between the two - they seem like the same reading with
a fixed offset between them (angle difference between the two values never
changes by more than 1e-06f as the headset moves around).
Posted on the aforelinked thread about this and asked about the known
motion detector init problem.
I have seen a bit of minor drift
on all 3 rotation axes but it always sorts itself out sooner or later.
Yaw drift is most noticeable when the pitch is extreme.
So far I haven't found any need for a re-centring option when using Z800.CHeadTracker::Calibrate (mis-named) only
applies to OptiTrack.
Added C3DVisor::KeepAlive which
keeps the headset awake.Don't call
this every frame.
Interesting - I'd got the
headset connecting properly every time, then I opened Fraps and ran the game; suddenly it can't connect to the
headset.
Unload Fraps, run again, connects fine.
Open Fraps again, connects fine.
Then it works, then it doesn't.... I can't see
a pattern >:(<-FIXED(6th)
3DVisor stereo: I need to be
able to run the game, in 800x600 debug, at its default settings or above,
with all the rest of the effects & stuff that I haven't even added yet, without ever dropping below 60 fps
(I've checked that the headset definitely doesn't switch to stereo when the
fps is below 60).
Currently I'm in the low 20s.No
amount of optimisation work is going to get me out of this (and optimisation
work isn't the best use of my time anyway).
So I'm going to buy a faster graphics card and have done with it :)
NB - in frame-sequential stereo mode I will get some speedup from only
having to do half as many updates, but that won't help much.
Didn't really do much work on
the game this evening, just a load of experimentation with the headset
(mostly logged)
4-Oct-2007
5.25
#revision 753
OptiTrack support is fully
working again, under the new headtracking system.
Installed a DirectSound example
project:
C:\Documents and Settings\user\My Documents\Visual Studio Projects\SoundFX
Had to move WDK include path to before DX ones, to avoid a
DirectSound entanglement.
added linker dependency:
EMADevice_DLL.lib
Synaesthesia:
Totally unrelated to anything, here's what 3.1415927 looks like in my
head.I gather everyone has
different colours, so this probably looks horribly wrong to anyone else.I'd be interested to hear from you if you
have the same colours:
#revision 756
Had to add post-build steps to
copy the appropriate EMADevice_DLL.dll to the appropriate output
folder.
Can now step into the source for that dll.
3DVisor.cpp: Because of the
funky overloaded 'EMADevice::operator new', I had a crash when the 'Z800'
object was a member of C3DVisor, when the construction failed.Works fine as a global.
Got the Z800 headtracking
working.
I can hardly describe how bloody
amazingly wonderful the Z800 headtracking is.Silky smooth, deadly precise, totally responsive.It's like it's not there.
After taking the headset off and putting the light back on, I find myself
looking around the room going "where am I"?
THAT's virtual reality!!
And I haven't even added the stereoscopy yet!Or the sound!
The 'headTracking\smoothing' option is now set firmly to zero, that's how
good it is.
Poor old TrackIR....seems so inadequate now.TrackIR support cost me about two weeks of
my schedule, too.
Did learn some useful stuff from it though.
TODO: try using the Z800's raw
compass data instead of its yaw value - apparently that works really well:
<-DONE(5th) I can't see any difference in results between using the yaw
and the compass reading, they seem like the same reading with a fixed offset:
Today marks one year since I
stopped drinking.
TODO: When I overhaul the devlog converter, I'll get it to spit out some
productivity stats that can be graphed (hours per week / month).
Should see that the past year has been significantly more productive than
the previous ones.The lack of
nights-out and subsequent hangovers will have played a big part in this, but
also the band activity has been sparse.
There was an important
development on the 14th or 15th of June - I suddenly stopped feeling regret
for the things I'd said & done in the past while drunk.This catalogue of memories had been cutting
and stinging me every day for nearly ten years, then suddenly it just wasn't there anymore.Very weird and very helpful :D
I have a feeling my short-term memory has suffered a bit, maybe as a result
of not being exercised by the constant retrieval of drunken memories new
& old.I seem to have to write
things down more than I used to, mais c'est pas grave.
Of course, for all I know these changes could have been brought about by
the Roaccutane rather than the cessation of booze activities; I wouldn't
know.
I probably won't start drinking again till after the project is
"finished", at which point lots of things are going to change.....
Found that there's an eMagin X800 3DVisor, which I gather is the
same as the Z800 but without headtracking.
I want to support this; therefore my headset-handling class is going to be
called C3DVisor rather than CZ800 (the two headsets use the same SDK).
Added:
- C3DVisor
(3DVisor.cpp&h) - takes care of all the 3DVisor specifics (headtracking
& other device features)
- COptiTrack
(OptiTrack.cpp&h) - all the OptiTrack specifics (covering TrackIR), ie.
most of the current contents of CHeadTracker.
Both these objects are members of g_game.
CHeadTracker now becomes a thin layer that pulls the appropriate
information from those two objects to produce standardised head transform
data.CHeadTracker also still handles
all the debug draw.
#revision 752
CHeadTracker::Calibrate: Removed
some half-written code that would try to auto-detect the OptiTrack camera
orientation.
Eek, I now have an
"OptiTrack.h" and an <optitrack.h>.Seems to work ok though; I suppose that's
what the angle brackets are designed for.
Added options category OptiTrack (g_game.options.headTracking.OptiTrack)
Moved the following options from headTracking into OptiTrack:
cameraOrientation
xHalfFOVTangent
yHalfFOVTangent
videoThreshold
Noticed occasional rain
occlusion failure outside House of Fraser (<-sounds a bit Westworld, that ;)
TODO: fix this<-DONE(12.4.8)
Got the game running with the
re-jigged headtracker code; isn't quite working yet
though.
TODO: finish this, add 3DVisor headtracking.<-DONE(4th)
2-Oct-2007
4.25
Installed the Microsoft Windows
Driver Kit 6000 (had it on a DVD at work, now copied to Installs
folder)
I can now build and run the eMagin example projects, yay!
FXC _DEBUG compile:
c:\Noctambule\Source\HLSL\EnhancedMesh.fx
Microsoft (R) D3D10 Shader Compiler 9.19.949.0046
error X3539: ps_1_x is no longer supported; use /Gec in fxc to automatically upgrade to ps_2_0
error X3539: Alternately, fxc's /LD option allows use of the old compiler
DLL
Moved the eMagin SDK to
C:\Noctambule\eMaginSDK, and put it into SVN.
Joined eMagin forums, username
'phil'
Been having difficulty
connecting with the Z800 head tracker.This is a known problem (TK3DERR_NON_DEVICE).<-FIXED(6th)
Some discussion:
BUT - disconnecting and
reconnecting the USB seems to fix the problem.
Hmmm....<-FIXED(6th)
What I'm going to do is port the
tracking code into the game anyway, then if it doesn't behave properly I'll
try initalising it from various different points in the game code to see if I
can find one that works reliably.<-FIXED(6th)
1-Oct-2007
3.5
v13.3, on 8800, has what looks
like fallback shadow junk (ERENDERTARGET_FALLBACK_SHADOW or
ERENDERTARGET_FALLBACKSHADOWREFLECTIONPOSITION or
ERENDERTARGET_REFLECTION_POSITION not being cleared, that's what it looks
like).
TODO: check it on quadro/PC375 again; try testing with rain occlusion
switched off.
It's quite possible that this will be fixed as a result of getting the
impact effects togglable (DONE).
#revision
744
<-got what looked like the
same problem here (now fixed).Needed
a zbuffer clear needed before generating the reflection image.
Made the clearing a bit neater now; it's now done at the start of
RenderSubset (rather than RenderEye).It's skipped if the render category is RAINOCCLUSIONDEPTH.Could alternatively add a renderData flag
for the zbuffer clear.
#revision
745
Rain splashes and rain ripples
can now be independently toggled without any glitches.
#revision
746
CBlockParser now correctly skips
lines commented-out with "//"
Renamed
ESTEREOSCOPYMODE_ALTERNATING to ESTEREOSCOPYMODE_FRAME_SEQUENTIAL
Renamed ESTEREOSCOPYMODE_DUALDISPLAY to ESTEREOSCOPYMODE_DUAL_OUTPUT
All the options enums use the convention EMYENUM_MY_VALUE
Elsewhere in the code, I also use EMYENUM_MYVALUE.Really I should've made a rule for this in
the coding conventions doc, but it's very minor.
27-Sep-2007
3.666
My VR
HEADSET arrived today!!!
We are now officially in Teh Futur.
Very pleased with it so
far.In terms of display quality, the
Z800 is pretty much exactly the way I imagined it.Fills about the same FOV as a monitor
(anything more generous costs thousands unfortunately).I haven't run anything stereoscopic through
it yet though; that's what I'm working towards now.
The optics are a bit blurry round the edges, but that
doesn't cause me a problem (especially since I was planning to add that
effect for this game anyway).<-loads clearer if
you ensure that the eyepieces are as close as possible.
It's also more contrasty than a CRT; dark stuff
gets a bit lost so I'll need to tweak my colour grading.<-tested, works well, see below.
The head-tracker looks very
promising too.Even just using its
default mouse emulation, it's working well.Currently looks like the tracking update rate is
a bit slow but a bit of smoothing will fix this.<-I
WAS TOTALLY WRONG, THE HEADTRACKING IS AMAZING, SEE 4TH OCT.
Using Z800 and TrackIR together is great,
except that the TrackIR's limited angular range suddenly becomes restrictive
when you start using an HMD rather than a fixed monitor.<-but the z800 headtracking is tons more useful than
TrackIR.
Being able to properly look around though, it's
so cool.I looked up at the rain
coming down around a streetlight, andhonestly felt like it was going to get in my eyes.I reckon if the game were set in summer
sunshine, I'd find myselfsquinting
when looking up at the sun as well.
Tried to build the Z800 example
projects.But they require the Windows XP DDK (driver development
kit), which is a 2.4 GB download >:(
download link:<-Got it now (2nd Oct)
TODO: download this beast and
hope it works!<-Got it now, and ti works (2nd Oct)
Joined the Microsoft 'Connect'
programme; this is necessary in order to download the DDK (also known as the
WDK).
Rain impact effects are now nearly safely togglable - TODO:
finish this!<-DONE(1st Oct)
DeferredLightingApply.fx now uses uniform parameters to toggle reflections,
anaglyph, rain splashes and mist.
Needed to use Catalyst control
panel to force 60Hz, otherwise the game display was horizontally shifted on
Z800.This probably means the game is
trying to use the highest possible refresh rate instead of the Windows
one.
TODO: fix this.
Z800 was coping badly with my
dark colours & subtle mist (coming out too contrasty), but thankfully
this is very easy to fix using colour grading.Tested with a curve that makes it look
somewhere between the original and graded CRT results (looks great now) :
Fixed CLockableTexture.OptiTrack debug image now displays again in
the calibration mode.
I think they just needed to pass unique resource names through to the
CTexture constructor.
COptions now displays a warning
message if it finds an unrecognised option category
(COptions::WarningUnrecognisedCategory).
Got CCleaner and Regmon.Fraps now works.
MOM.exe is "Catalyst
Control Center: Monitoring program"
Started having a look at
performance.Need to stay above 60fps
at 800*600 in DEBUG mode, in order to debug in Z800 stereo mode.......
I'm a long way off this at the moment (as low as 45 with 0 shadow projectors).Lighting seems to be a real killer;
lighting bounds are too generous in many cases.
The easiest fix would be to buy a faster graphics card ;)It's tempting....
TODO: finish getting rain impact
effects safely togglable (see #error)
TODO: speed up the ground
collision by not including small polys in the cache (kerb bevels etc!!)
Alternatively, set up automatic generation of collision geometry - that
would be better.
Profile 'Phil.txt' is now always
loaded when running from the debugger, even in Master config.
25-Sep-2007
2.75
Tracking down the patchwork
normals bug:
- not a change in Patchwork.fx
- patchwork (TS) normals themselves are fine; must be a bug in ground.fx
<-FIXED.Ground.fx revision 569,
lines 703 [& 704]: basically a remout of a required transformation for
the normal.Variable 'normal' at this
point is not the WS normal that needs to be output.A bit of renaming might have helped there.
Found that the headtrack
calibration mode was crashing since it changed over to use
CLockableTexture.That means 13.2
would crash if you pause when using TrackIR.
As a temp workaround, the debug image is not drawn in v13.3 (but the debug
meshes etc are).
TODO: Fix CLockableTexture!!!<-DONE(26th)
NOTE: WOWvx mode's CLockableTexture is working
fine.
below: I've been thinking
"Zombies?... Spiders?..." for quite some time now.If only I'd thought to put the two
together!
NB. this is not my enemy design, this is from Dead Space.It's exactly the kind of thing that I'm
after though - not just scary, but grotesquely and instinctively scary.
#version
13.3
#revision 740
Released version 13.3
24-Sep-2007
2.25
Orthographic culling now allows
for a sheared view matrix (see UpdateOrthographicProjection).
This prevents objects popping-out from the rain occlusion depth image.
Added define
RAINOCCLUSION_DEBUG_DEPTH_IMAGE, since it was a bit fiddly otherwise to get a
debug display of that depth image.
EEK!Patchwork normals are wrong (unless you
look south)!When did that start?TODO: FIX THIS!!!
- Wrong in 13.2
- Wrong in 13.0
- Fine in 12.0
- got broken between 19th June & 10th September (revisions 625 to 721
inclusive)<-FIXED(25th)
23-Sep-2007
1.8
TODO?: FX: make a header
function for common rain occlusion stuff.Currently there's some duplicated code+constants+sampler.
Ground.fx: ran into the vs_3_0
output register limit (again) trying to add rain occlusion, then noticed
something that should've been obvious before:
In the PS, I should be (and am now) doing this: input.tangent =
normalize(input.tangent); input.normal =
normalize(input.normal); float3 binormal =
cross(input.normal, input.tangent);
not this: input.tangent =
normalize(input.tangent); input.binormal =
normalize(input.binormal); input.normal =
normalize(input.normal);
Removes a VS output, reduces the required vert size, saves a normalise in
the PS.
Checked that the binormals are still the right way round, by looking at the
road text (this shows an obvious wrongness when the binormals are flipped).
ShaderBuild.exe: removed fxc
logo text (/nologo) - makes it nice & easy to spot errors.
Compiling individually still shows the logo text (as intended).
Ground.fx: all ripples (flow,
impact rings) are now applied using the 'lerp TS XY' method (previously
lerping TS XYZ).
TODO: re-tweak ground normal/ripple strengths<-nah it's looking ok now I've fixed the normals bug(25th)
Ground.fx: Rain impact ripples
now have rain occlusion.
TODO: characters need to be able
to draw into the orthographic depth image.Currently they don't seem to manage.
The areas of shelter are
starting to feel sheltered (but audio will help this enormously).You get a nice contrast when you step back
out into the rain.This is good!
22-Sep-2007
7.55
Strengthened the falling rain so
it looks denser and glintier.
It's now visible even against dark ground or dark sky.
Profiles: integers can now be
used in place of enumeration values (CBlockParser::ReadEnumeration)
Spent quite a while
un-knackering the fallback shadows.Some weirdness:
I'd been trying to use reflection position Y values less than a certain
height to mean 'no shadow'.But it
seems I can only get this to work if FallbackShadow.fx checks for an exact Y value rather than
greater/less than (which I'd be far happier with because it's more flexible
towards RT format).
See FALLBACKSHADOW_NOSHADOW_Y.
Rain occlusion now shears
correctly.The shear is pivoted around
the viewpoint height.
TODO?: improve orthographic visibility testing to
account for the sheared view matrix?This will only be worthwhile if the occasional popping-in&out of
rain occlusion around objects becomes visible.I haven't found it noticeable so far.<-DONE(24th)
Rain occlusion volume's XZ
bounds are now clamped around the bounds of the visible RainFalling
cells.
That's why the depth image appears to jump around all over the place when
you debug it (just in case I forget next time I look at it).
Remembered about the
'vpRelativePositionIn' parameter of TestSphereVisibility.
#revision 735
Rain occlusion depth image is
now generated at start of the frame rather than per eye.
TODO: deal with frame-sequential stereo
Re-enabled linear mag filtering
for the mist - loads better!Must've
disabled it as a temp test?(DeferredLightingApply.fx)
#revision 737
Rain occlusion now creates a
single matrix constant for RainFalling.fx (vpRelativeWorldToUVW).
RainSplash.fx: Removed unused
(remmed-out) pass 3
#revision 738
RainSplash now gets the same
occlusion as RainFalling.The
occlusion bailout is on pixel shader pass 2 (the more expensive one, with the
loop to extrude the splashes).
It's quite nice now to see the boundaries of the RainSplash effect waving
back & forth as the wind changes.The audio wants to match this so that when you're in shelter you hear
the patters waving in and out (or at least, you think that's what you hear).
21-Sep-2007
3.33333
At work, made public the code
and graphics for RainFalling, Mist and VolumeLighting.I was reluctant to do this, but I would
have appeared unhelpful otherwise.
Release log: documented the
workaround for the v13.2 WOWvx startup crash (having noticed that people from
the Netherlands had spotted the WOWvx support ;)
Tested the workaround as documented.
Improved profile Example_3DDisplay_Philips_WOWvx_42inch.txt :
Doubled the dimensions of all RTs except the reflection one, to compensate
for the half-dimensions frameRT that gets stretched over the whole screen by
the display hardware.
As I'd assumed, Z800 stereoscopy
works with any graphics card & drivers if the application generates the two views and sends them
frame-sequentially.Phew!
The game is now explicitly
v-synced (D3DPRESENT_INTERVAL_ONE), in both windowed and fullscreen
modes.This is necessary for
frame-sequential stereoscopy (besides, who wants tearing?)
The D3DXUT defaults previously only enabled v-sync in in full-screen mode
(using D3DPRESENT_INTERVAL_DEFAULT).
Rain occlusion: Decided to avoid
the need for a depth surface, by using these renderstates (below) for the
alpha-only depth image.Could have
borrowed the projector depth buffer but decided not to - keeps things simpler
and neater and more flexible.And I'm
guessing any speed difference would be negligable.
The important bit is the MIN alpha blend:
CULLMODE = CCW;
ZWRITEENABLE = FALSE;
ZENABLE = FALSE;
ALPHABLENDENABLE = TRUE;
SRCBLEND = ONE;
DESTBLEND = ONE;
BLENDOPALPHA = MIN;
COLORWRITEENABLE = ALPHA;
below: FallingRain.fx now has
working occlusion.Currently
it's just vertical, but next I'll shear the depth image's view matrix
according to the angle of the rain.<-DONE(22nd)
TODO: somehow use the rain occlusion depth image to reduce the mist in
sheltered areas?
20-Sep-2007
3.43333
FX:
"ID3DXEffectCompiler::CompileEffect: Only 1-d shader arrays
allowed"
Hmmm, the rain streak pixel
shader now knows where each pixel is relative to the surface of the
environment.This info could be used
to produce a new, more accurate rain splash effect......
TODO: investigate this!
Resurrected and tidied the
orthographic stuff in CViewInfo.Rain
occlusion depth image is now working properly.
Occlusion of FallingRain.fx is
nearly working; not sure at the moment that my WS occlusion positions for the
streaks are correct (see fx error).
19-Sep-2007
3.93333
Back to the weather
effects:
Rain occlusion, because
it's a bit shameful that there currently is none.
Decided that the occlusion
system for the rain visuals will not need to deal with audio at all.
The audio effects of sheltered areas will be triggered by bounding
boxes.Boxes rather than rectangles /
fences because I want the sounds to work properly with arbitrary camera
positions.
Added CRainOcclusion (owned by
CModeEnvironment).Considered making
it a CProjector subclass; decided it would be better not to.It doesn't really tie-in closely enough
with any of what I want for the lighting (I don't have any use for
"directional" lights, as far as I know).
CRainOcclusion currently borrows
the floating-point alpha channel of the fallback shadows' reflection position
RT.
Fallbacks' yes/no signal is now in the Y channel (see comments).
v13.2 crashes on startup if
WOWvx mode is enabled.I must've
forgotten to test WOWvx after
changing all the naming (13th).This
is fixed for v13.3.
From now on, I'll remember to clean the Master config of Shaders.sln before doing a build - this
catches a few potential problems to do with shader files.
Retired a few old FX files into
the 'graveyard' folder.
below: interesting
accident.It flickered like lightning
or like an old film.
Made a good start on the rain
occlusion.Found that the orthographic
CViewInfo stuff has been kinda half-removed, so I need to resurrect that
before the depth image can render properly (see #error).
#backup
Backed-up the project.
18-Sep-2007
2.5
Fixed the WOWvx header EDC
- an alteration to UtilCRC was required (ie. UtilCRC must not have been
conforming to the CRC-32 standard before)
- I had to reverse the EDC bytes.Presumably I had to do this last time as well? - check this
Got the header displaying
again.Had to write the bytes to the
texture in BGRA order, but the texture format is D3DFMT_A8R8G8B8.I don't know why the component order gets
swapped like that.
Current header (matching the 'standard' video header) is shown in blue on
the header map
F102408000001F3A7B38 :
TODO: tweak the WOWvx settings
in Default.txt; include the tweaked values in v13.4.
Anaglyph ghost reduction is
currently hors service.Added a
warning message about this (if the option is selected) and stopped it
crashing.
Updated the example profiles to
use the new stereoscopy options.
#version
13.2
#revision 727
Released version 13.2, which
contains the following improvements to WOWvx support:
- smooth viewing zone transitions by default.
- all 3D display parameters can now be configured using options profiles.
16-Sep-2007
Got loads of nice rain &
weather recordings from S, so I'll be able to start experimenting with audio
shortly, which should add loads.
It'll involve a lot of guesswork though, because my right ear is
practically stone-deaf to the high-frequency sound of rain :(
Some of the MP3s decode with
nasty clicking artefacts though, and I've heard this problem before.What am I doing wrong there?I'll try opening them at work and see if I
get the same thing.
13-Sep-2007
6.333
#revision 723
Made a lockable texture class
(CLockableTexture, LockableTexture.cpp&h), in preparation for generating
the WOWvx header in code.
Inherits from the CTexture resource, but NOTE: doesn't use m_pResource
(neither of its two textures can use D3DPOOL_MANAGED because they're both
D3DUSAGE_DYNAMIC).
Changed CHeadTracker over to use CLockableTexture for its debug image.
The reason I want to generate
the WOWvx header in code is that it's currently using 'signage' mode rather
than 'movie'.
This means the view transitions aren't smooth.And I really do want this game to be a solid
demo of the WOWvx displays.
Also, it'll be good to have all the parameters adjustable, just in case.
#revision
724,
725
Tidied-up the stereoscopy
options:
- the term '3DTV' is now changed to 'WOWvx' in both the options and the
code - been meaning to do this for ages.
- added option categories WOWvx and anaglyph (moved the relevant stereoscopy options into these).
NOTE: the new sub-categories aren't nested in the profiles, but they are
nested in COptions.Could improve the
parsing to handle nested blocks later.
Release log: I'm now
striking-out fixed bugs.This means
that when you look back at the release notes for an old version, you can
easily see which of its bugs are now fixed.
Added a 'new files' log which
should make the build process a bit smoother (usually I forget about the new
files I've added and have to build several times before it runs) :
Got the WOWvx header texture
generated in code from profile options.
TODO: Fix this - currently the EDC value from UtilCRC
is wrong.<-FIXED (18th)
TODO: need to squeeze this many
hours in every night! ;)
12-Sep-2007
1.666
v13: The ground patchwork decals
appear mis-sorted on PC375 (7800? 7900?) and PC558 (Quadro).
<-FIXED in v13.1.I had been
re-using landscape\dimpleNormalsRT (now 2TTT) for the dimples height image
(which is formed using blending).The
effect on SM3 nVidia cards was as if blending were disabled in
PatchworkDimples.fx.
Now using non-loading member landscape\dimplesRT, which copies
dimpleNormalsRT but is forced to D3DFMT_A8B8G8R8.
#version
13.1
#revision 722
Released version 13.1.This just fixes the dimple height blending
on cards that can't blend to 2101010 render targets.
10-Sep-2007
Grabbed v13 videos and put them
on YouTube.
NB this was an entire evening's work (not logged).Next time round it'll be a bit easier
because I'll be able to cut & splice the bits I like rather than trying
to record perfect clips.
Something that was made very
clear by this capturing exercise was that my shadows need fixing quite
urgently.I had to throw away loads of
recordings just because of shadows breaking-up.They wasted a fair bit of my time.
9-Sep-2007
11
TODO: for version 13:
- get blurQuality working<-DONE
(8th)
- fix POM edge glitch (just a bit of limiting
needed somewhere?)<-done (9th)
- get OptiTrack calibration mode usable<-DONE (9th)
- get options saveable<-nah that's for the next build
- finish motion blur?would be nice, shouldn't take too long<-nah that's for the next build
- tidy OptiTrack detection etc<-done
- finish the headtracking effects<-no, need to get v13 out today.v13 will have 'rudimentary TrackIR
support'.
- fix dimples (view tilt)<-DONE (9th)
- check occlusion culling (view tilt)<-gave it a quick check and it looks fine
Velocity.fx: disabled lerping of
each texel towards its maximum velocity sampled during the blur.This was making itself too noticeable - eg.
running towards a lamppost, the lamppost would be bobbing up and down and
this movement would vertically blur the background around the object.
#revision 715, 716
Fixed the POM edge glitches, by
adding two limit values:
- min tangent-space view Z position (currently 1 cm), this also prevents a
possible db0 (MIN_TS_VIEWPOINT_Z)
- max parallax displacement (currently 10 cm, should be plenty) (MAX_PARALLAX_LENGTH)
Fixed the dimple problem when
the view was tilted.This yielded lots
of improvements:
- dimple processing now draws using a ground-plane quad instead of a very
approximate ground-covering screen-aligned quad
- was therefore able to chuck out a load of uploaded constants, and
calculations on both CPU & GPU sides (see
GetViewSpacePositionOnGroundPlane in h_Global.fx)
Weird, after making the changes
above, had to roughly half PATCHWORK_DIMPLEPROCESSING_SMOOTHNESS.Can't see why that would be.
Ground.fx: Cheapened the dimple
normals influence by using the 'lerp TS XY' method, as I've been calling
it.The strength can now be adjusted
using GROUND_DIMPLE_NORMALS_INFLUENCE (better to use this than PATCHWORK_DIMPLEPROCESSING_SMOOTHNESS,
which is now set to an optimal, bumpy as possible, value - to keep the normal
colours strong).
#revision 718
The above change partly
alleviates the refraction-vs-POM glitch visible around road paint (see 31
July 07)
HDRLighting.fx: added
COLOUR_GRADING_STRENGTH, because it's slightly quicker to tweak this than the
ramp texture.Might be temporary;
definitely neater to edit the texture.
Dimple normals now use a 2TTT
target - why didn't I think of that before? (don't need to blend to it; alpha
channel not needed because only the XY of the normals is used; blue channel,
instead of alpha channel, now holds height).
Looks much nicer now!
Also added
GROUND_NORMALS_FINAL_TS_Z_MULTIPLIER (Ground.fx), because the patchwork
normals were looking a bit lost now that they're lerped towards dimple normals rather than rotated onto them.
Added an ambient colour to the
volume lighting image - definitely a good idea because some of the fainter
mist & rain had got lost under the colour grading.
Ambient colour is defined by LIGHTMANAGER_VOLUMELIGHTING_AMBIENCE.
Removed hardcoded ambience from RainFalling.fx
Targetting now takes into
account head-tracking.
Setup: Added the three OptiTrack
DLLs to the system folder.
NOTE: These are sourced from C:\Program Files\NaturalPoint\Optitrack\bin,
but I've also put backups in the Builds\OptiTrack folder.
#version
13.0
#revision 721
Released version 13.0.Didn't have time to do new videos, will do
this tomorrow night.
8-Sep-2007
1.333
Oops, been distracted by work,
YouTube and band stuff.
Some YouTube-related name-dropping:
- Mark Stirton (of Stirton Productions) said the game looked interesting
(and weird)! :DAce!
- Mario Cavalli kindly posted-up his 1992 film Soho Square for me!(this is very relevant to some ideas I've
been brewing-up for the mythical "next project" that I'll never
have time to do) :
Found that my motion blur had
never been adapting its #samples per pixel to the pixel's velocity - it was
always using the max #samples.
Can't coax the HLSL compiler into letting me do this - has anyone else
managed? :
- sample value X from a texture
- use X to determine the number of samples to take from another texture
So for now, I'm going to let the
Wookiee win and just always use the maximum number of samples.When I next upgrade DX, I'll have another
try, since I have a feeling there was some compiler bug mentioned in this area.....
Got motionBlur \ blurQuality
working.The 'max' setting (not
recommended!) uses 169 samples - that's literally the maximum number I can
use before it fails to draw the effect.
2-Sep-2007
1.6666
NOTE: C++: can't check enum
values at compile time :(
C++: Why does this never
work?It's straight out of an MSDN
example but it doesn't work.What am I
doing wrong? :
#pragma warning( push )
#pragma warning( disable : 4100 )
// Some code
#pragma warning( pop )
Headtracking: no longer crashes
on startup when no camera is found
Started getting option
'motionBlur \ blurQuality' Working; ran into a bit of compiler fussiness - todo: finish this<-DONE (8th)
1-Sep-2007
8.93333
I'm finding that motion blur
mishaps create some really nice dream/nightmare effects.
TODO: I reckon hijacking the motion blur effect (random rippling in the
velocity buffer?) will be a great starting point for the dreamy look.Not entirely unlike the Batman Begins
hallucinations, which I really like.
Ooh, If that motion blur distortion can be masked, it could be great to use
selectively on stuff like the enemies or the environmental text (the stuff
that I would *want* to look closely at in a dream would but struggle to get a
clear image of).
below: a couple of other motion
blur mishaps.Crashing waves on the
left (if you blur your eyes); glass/gem overlays on the right.Glass has been one of my ideas for the
menus etc so that was a nice coincidence.
If I do use any glass/liquid for the menus, it would be a good idea to have
it reflect the environment - this would show the player their own reflection
which would be neat.
Found that my motion blur
velocity image had been one frame behind, now fixed;
Spent most of the day trying to
find a good processing technique for the velocity buffer.This is to deal with velocity
boundaries.Tried loads of stuff, but
in the end it became clear that the 'clever' methods need loads of loop
iterations (far too many) to actaully give a good result.
So I'm now just doing a simple blur on the velocity, plus
lerping each texel towards the maximum velocity sampled during the blur<-DISABLED(9th, was causing bad artefacts).Works well enough.
1-Sep-2007
6.08333
Added option headTracking
\ smoothing (1-t value for lerping to the head
transform each frame).
Can't be bothered making the smoothing framerate-independent [yet].
Tested option headTracking \
cameraOrientation.
Added Light.cpp&h to SVN
(gulp, these were unversioned till now).Checked that these were the only ones.
Searching for free rain sounds
online seems to be a complete waste of time.The loops need to be ultra-crisp or they're no use at all.
Added these options and got them
working:<-TODO: get blurQuality working<-DONE(8th)
motionBlur \ blurQuality (replaces maxSamples, because my blur loop needs to be
unrolled - compiler bug?)
motionBlur \ blurVelocityImage (blur the velocity image to prevent edge artefacts at velocity
boundaries)
motionBlur \ distanceForMaxSamples (step distance above which the maximum number of samples is
used)
Options: Unrecognised option
warnings now display the name of the profile in which the problem was
encountered.
30-Aug-2007
Some very good (and completely
unaffordable) HMDs, using 24 SVGA (800x600) inputs:
The diagonal FOV I get from this
monitor (LG 700S) is about 60° (35 cm from the screen)
Got a sudden blue-screen crash
(trying to open Calculator, with lots of Explorer windows open)
Something about NTFS.SYS
When I got back into Windows after the restart, it thrashed the hard drive
for a couple of minutes before doing the usual startup stuff.
NOTE: lighting tails are (and
always have been) needlessly fat for lights close to the ground.
TODO: proper specular highlight shape calculation, in conjunction with
CLight::ClipTailBelowHead
TODO: dimples are glitchy in the
distance when tilted; fix this
Got the lighting geometry
working with head tilt (just needed a load of aspect ratio compensation to go
with therotation).
Didn't do anything special for ground slope though.No point adding that till it's really
needed.Mild slopes won't cause
visible problems.
Decided not to attempt
auto-detection of headtracker camera orientation - I've lost enough time to
feature-creep recently.
Instead, added option headTracking \ cameraOrientation
TODO: deal with
CLight::s_handleTilt
22-Aug-2007
2
Added function Vec2Average
Downloaded VirtualDub, for
preparing video files
Trying to get that lighting
geometry tiltable.....<-DONE (28th)
21-Aug-2007
2.5
Got the shadows working with
head-tracking (I had forgotten to check the 'useHeadTracking' flag, on the
projector camera).
Reflections are fine (UtilDrawGroundCoverQuad seems to work ok when tilted;
could probably optimise according to the tilt though).
Got the mist working with
head-tracking.
TODO: need to improve lighting
lollipops so they support view tilt (and ground plane
slope, while I'm at it).
Easiest thing to do will be to keep the existing calc of the head shape and
(from the clipped head shape) screen area, then rotate the head and plot the
tail according to the screen-space vector from the light pos to the reflected
light pos.<-DONE (28th)
Intro: TODO?: Use a floaty
flying camera (running along a spline) + phospheney blurry effect?Flying around with head-tracking looks
great.Phospheney blurry effect would
be perfect for the edges of the screen, if it wasn't too annoying.
TODO: Should really give some
thought to using "dream text" as the theme for the
menus/front-end/HUD.....
Just need to make sure it doesn't end up messy, with all the stuff going on
in the background as well.
It's tricky because when you're tweaking graphics options you probably
don't want the scene to be obscured or filtered in any way.....
Would be good to get all the menus etc started SOON.
21-Aug-2007
2.91667
Added function MatrixLerp
Added some head-track
smoothing.Currently simple lerping.
Motion blur now takes into
account head-tracking.
You can actually peek round
corners now that the head-tracking is in, it's cool!
Started getting the shadows etc
working with head-tracking, TODO finish this (see #error), TODO: tidy/rename
stuff
TODO: try adding head-track
steering method
Found that generating the
headtrack debug image is quite costly; TODO: better control for when this is
& isn't generated.
20-Aug-2007
Overtime.Wait...didn't I make a rule about that?!
19-Aug-2007
10.25
CMeshFrame::RenderIndividually
no longer tests visibility in non-_DEBUG builds
NOTE: any kind of split-screen /
picture-in-picture features are now officially not going to happen (time is
getting short so I need to ban as many features as I can - a 'to-not-do'
list, for a change!).
For the head-tracking, I needed
to overhaul the camera / viewInfo system a bit:
- cameras are no longer involved with stereoscopy
- CCamera::Apply is now CViewInfo::ApplyCamera (this applies the offsets
for head-tracking and stereoscopy)
- cameras are not aware of head-tracking
NOTE: in anaglyph modes, motion
blur setting currently has no effect (no blur).May or may not get the two working together
later on (would be nice, of course).
Got the anaglyph working again
(had been broken by the motion-blur) : LightManager.cpp
(below) Checked that WOWvx mode
hadn't got broken by motion blur or colour grading.And no, I'm not going to motion-blur the
depth data!
(NB - this is not the aspect ratio used in WOWvx mode)
MaxScript: head.max:
head->clip matrix:
$trackclip.xform.gizmo.transform
...but I can't get that matrix to work in the game.Certainly had to negate all the X
components.......?
Added CModeHeadTrackOffset
(ModeHeadTrackOffset.cpp & h) - the game mode for setting the head/clip
offset for head-tracking (lets the player move & rotate the head model
into position relative to the trackclip).
Pleasingly, this took no time at all and worked straight away.
Got the player's head transform
correctly tracked ^_^It's currently
working just like the TrackIR support in MS Flight Sim (ie. 6DOF and suitable
for a head-mounted display).
TODO:
- Some kind of smoothing, because the low camera res
really shows now that it's controlling the view.
- handle the slight glitch where the sphere intersection in
CHeadTracker::TestZPosition fails (z pos close to max z).
- make the tracking clever enough to understand when the point order
crosses over.The TrackIR 4 software
doesn't do this, but it would improve angular range quite a bit.
TODO: fix shadows &
reflections, which look like they're still using the main camera position rather than the viewInfo position<-fixed (21st)
TODO: the 'virtual window'
head-tracking effect
18-Aug-2007
6.583
Planned the head-tracking
calibration.Found that it can work by
requiring the player just to look at the screen centre (rather than the
camera as well) provided the following assumptions are true:
1. the camera is known to be mounted somewhere on the screen plane, OR the
distanceToScreen setting is reliable;
2. the player's 'head' position (point between the eyes) lies on a line
perpendicular to the screen plane and originating at the screen centre.
An assumption I definitely don't want to make would be that
the camera is positioned at the screen's X or Y centre position, because I
know people like to put it on the left hand side of the monitor top for
example, to get a better view of the trackclip.
TODO: in calibration, try
interpreting objects in each orientation mode; see which one gives the most
sensible head orientation & distance.<-NO, MORE
COMPLICATED THAN IT SOUNDS + FEATURE CREEP! (28th)
Added typedef QSORT_CALLBACK
Swapped EHEADTRACKEROBJECT order
(now top, middle, bottom), less confusing
Got the binary depth search
working - very effective.
I've now put my camera quite far
away, slightly behind the screen plane and way off to left of the
monitor.
This gives really good coverage, so I want to support this kind of
unrestricted camera positioning.
Tried to borrow the TrackIR
calibration head model, an .X file; couldn't get it to display as anything
other than a crushed-up blob though.Is it skinned or something?Appears fine in the DX viewer.
Made a quick head model for
tracking debug
16-Aug-2007
2
Added some v13 preview videos to
the game's new Bebo page.Feel free to
'add' me:
Made the headtracking simpler
(cheaper) and more stable most of the time, by using the vector between the
bottom & MID lights as the trackclip's 'up' axis (previously bottom &
top).
Started changing over to a
binary search for the clip's Z position, TODO: finish this, too tired now
Re-learned how to do
simultaneous equations - needed to happen sooner or later.
(below) Got some very
official-looking TrackClip PRO measurements from the OptiTrack forums; used
simultaneous equations (arc-arc intersection) to derive coords from these
(now using these coords, and the tracking does look more stable as a result).
NOTE: there were more significant figures in the trackclip measurements in
the first version of that post; I used the first version, because those
numbers must have come from somewhere good.
Corrected the trackclip model to
match the new light coords.
14-Aug-2007
1.5
Fixed all but the most minor of
my tracking glitches.I had just made
a couple of really clumsy mistakes, causing code to be in the wrong order and
variables not to be initialised.Must've been sleep-coding again!
It's now nice & stable.The only remaining glitch is when the
sphere intersection in CHeadTracker::TestZPosition fails (z pos close to max
z).Shouldn't be too hard to fix that.....
Hoped that
GetDistanceFromRayToRing would completely solve my tracking problem, but it
didn't........ need to investigate further.......<-working
now (14th)
Currently getting totally
bizarre results (see #error). <- fixed 14th
12-Aug-2007
8
TODO: see flight sim forum
below, where someone suggested the same approach as mine for using the
TrackIR (the projection adjustments were even mentioned) :
tracker fov: another possible v
fov is 2*atan( tan(46/2) * (3/4) ) =
35.31838401289530331635147238541
measured the TrackIR 4 camera's
fov:
at 50 cm:
X extent is 19.75 cm
Y extent is 16.35 cm
aspect ratio: 1.2079510703363914373088685015291
X half-fov tangent: 0.395
Y half-fov tangent: 0.327
X half-fov degrees: 21.554019920478842643648210638558
Y half-fov degrees: 18.107744742279231622861032074031
Double-checked the measurements
above by using x=0.5 and y=0.5 positions; all bang-on, no sign of barrelling
/ fish-eye.
Made some handy lightweight
metre & half-metre sticks by chopping-up an old tape measure.Why didn't I think of that before?!
Got a fault in my vector
tracking at certain angles, but I'm pretty sure I know what's going wrong
now, at last.It's because I avoided
figuring-out GetDistanceFromRayToRing this morning, and used a different
method instead, which I now know to be incorrect.
GetDistanceFromRayToRing is trickier than it sounds...Hopefully fixing that will put everything
into place.
<- fixed 14th
Added GetDistanceFromRayToPoint
11-Aug-2007
8.25
cross-product: swapping the
parameters negates the direction of the output
Added axis vectors &
negative axis vectors.Obviously it
would always be better to optimise these out, but they are still handy to
have around.
Put 2 XForm modifiers on the
trackclip model, so that middle.x = 0 and top.xz = 0
above: vector tracking with
TrackClip Pro is now working well :D
The depth calculation is
iterative (because I'm not clever enough to do it the fancy way, assuming
there is one), but it's all good and doesn't seem to need as many samples as
I'd imagined anyway.
TODO: use a binary search for the depth.<-done (18th)
Asked on the NaturalPoint forums
if they could provide exact measurements for the TrackClip and the camera
fovs
7-Aug-2007
2.4
Tracking: tried prioritising my
'objects' (points) by rank, and by score, but seemed to get unhelpful results
either way.Maybe it's because I have
to set up preferences for min/max/ideal object size first.
For now I'm prioritising by area, which gives the result I want.
Added Vec2Mul and Vec2Div
Changed the VEC2 functions to take pointers rather than references, for
consistency
Tracking: Note: object X coords
are negated, no idea why
(below) The trackclip model is
now pinned neatly onto the lowest object (follows it cleanly round the
screen)
NOTE: model is currently facing away from the camera; when all 3 points are
tracked, it should (fingers crossed) swing round to face the camera.
TODO: get the top point pinned next
6-Aug-2007
3
Continued setting up vector
tracking & its associated debug rendering.
TODO: use the qsort to discard the non-useful objects
(rank? score?)<-using AREA
TODO: fit the debug image to the display width
Some free sound samples
(potential monster noises etc).Need
to find more pages like this:
Another use for the
head-tracking: Head lamp.Will be
great to see this picking up subtle head movements, will make it way more
real.
Because a head lamp doesn't really require any player animation, I can (and
should) put this in pretty soon.Will
be useful for inspecting unlit areas.
A 'light cannon' or flare gun would be useful too.
Added HUDObject.fx for the
trackclip model.This might well be
used for weapon icons etc.
Corrected the measurements of
the trackclip model.Now 100%
confident about the side view (drawn at the screen plane, it's perfectly
'actual size', with the 3 light positions matching, as well as the numbers
all being nice & round).
The X offset of the mid light is currently 1.25cm, which may or may not be
correct; tricky to measure.
Got the trackclip model
displaying properly.
4-Aug-2007
13
TRACKING: XYZ coords seem to be
in millimetres, times two, the coords of the bend of the vector clip.
When X & Y are 0,
Z is 520 when the bend of of the clip is 26cm from the camera.
TRACKING: running around 50fps,
I'm requiring 197-199 loops to pick up a frame from the camera.
Running at a bad framerate, this number jitters between the discrete values
of 1 and 199.
Removing the per-frame LED commands changes this only slightly.<-this stuff's all variable
OptiTrack: Been getting vector
tracking results that don't make sense (especially for X & Y
position).The values move in the
right direction but apart from that they don't make sense.eg:
Procedural texture info: Search
for "Using Dynamic Textures" in DX help
Un-included wxdebug.h, for
reasons I can't be bothered describing; then had to add my own ASSERT and
REMIND macros (global.h)
From DX docs:
"Use square textures whenever possible. Textures whose dimensions are
256x256 are the fastest"
My camera's FOV is quoted as 46
degrees (I think that's horizontal)
So until I've done some measurements, I'll assume that the half-fov
tangents are in the same ratio as the res, giving me a vertical fov of 38
degrees (38.248394107273148325087524587558)
trackclip: 11cm from top to
bottom LED
Built a model of the trackclip
to help me with debugging the vector tracking.
Got the camera's image
displaying in the correct aspect ratio in front of the view.Will use this as a background for my
trackclip model.
Took ages to get the image into a texture and then onto the screen.Had to use a D3DPOOL_SYSTEMMEM texture for
the locking, then UpdateTexture onto a D3DPOOL_DEFAULT texture so it can be
rendered.Trying to render the
systemmem texture always gave white.Locking the default-pool texture didn't actually fail, but gave
corruption.Both textures are
D3DUSAGE_DYNAMIC.
3-Aug-2007
8.75
Spent all day trial-and-erroring
with the basics of the head-tracking, which was jerky and weird to begin
with, because I wasn't retrieving the frame data in the right way (amongst
other things).
It's now smooth as silk, but as far as I can tell, OptiTrack's vector
tracking only supports the cap, not the trackclip :/
I don't think I can justify the time it would take me to work out the
TrackClip tracking myself.Asked on
the OptiTrack forums if anyone knew how to do this.
Threads before OptiTrack init (note all the IDs change from run to run) :
1976
__freeCrtMemory (main thread;
dunno why it has that name)
Normal
0
744
Win32 Thread
Time Critical
0
3064
Win32 Thread
Time Critical
0
2940
Win32 Thread
Time Critical
0
pCameraCollection->Enum adds the following threads;
pCameraCollection->Release removes them with the following output
The thread 'Win32 Thread' (0x54c) has exited with code 0 (0x0).
The thread 'Win32 Thread' (0xb14) has exited with code 0 (0x0).
The thread 'Win32 Thread' (0x1a4) has exited with code 0 (0x0).
The thread 'Win32 Thread' (0xe40) has exited with code 0 (0x0).
3212
Win32 Thread
Highest
0
2244
Win32 Thread
Normal
0
3896
Win32 Thread
Normal
0
2404
Win32 Thread
Normal
0
292
Win32 Thread
Normal
0
pCamera->Start
adds the following threads; pCamera->Stop removes them with the following
output
The thread 'Win32 Thread' (0x588) has exited with code 1 (0x1).
The thread 'Win32 Thread' (0xcac) has exited with code 35 (0x23).
1508
Win32 Thread
Normal
0
3040
Win32 Thread
Highest
0
WEIRD: OptiTrack shutdown,
probably because it removes threads or something, can't be done in a
destructor.If I try to do it in
~CHeadTracker, there's a crash when it exits the function.
Now using Init & Shutdown methods instead.
TrackIR: To view profile
XMLs:
- remove '?' from start and end of XML tag
- add </xml> at end of file
TrackIR: noticed that pitching
movements were causing a slight roll to be read as well.This was because I'd angled the camera, on
the side of the monitor, to point inwards so as to centre the lights.Aligning the camera with the screen plane
fixes it, but gives me less rightwards truck range.
TODO: at the centring stage, fix this problem up
From the OptiTrack forums,
preferred object size:
"For the Track Clip Pro - on average i would say between 75 and
100."
Fixed typos in
Example_SpeedyLowRes.txt, causing wrong display dimensions.
2-Aug-2007
3.83333
Ported basic 6DOF tracking code
from the OptiTrack forums (link below).This was a butchered version of the command-line sample, with the ATL
stuff taken out.Had to add pFrame->Release after pFrame->Free, otherwise it would
crash after 19 or 20 updates.
Might've also had to frame=NULL before each camera->GetFrame, can't remember.
Got a rough test whereby the
player's camera is driven from the CHeadTracker instance.Pitch & Yaw seemed sensible, but when I
added roll....currently looks like it might be trying to read a track-cap; or
am I maybe just doing something wrong with the way I'm combining the
rotations? TODO: FIX THIS
OptiTrack command line sample
initially didn't compile (unicode entanglements).
Got it compiling by removing #include
<atlbase.h> and all the ATL usage: CComPtr
(a type of smart-pointer), CComVariant (a VARIANT wrapper).
Then just had to set frame=NULL before each camera->GetFrame.
My camera doesn't support
greyscale imaging :(
Would've been awfully groovy to project the player's face image into their
in-game face in real-time.Instead,
could take in a web-cam stream (this would require a calibration mode
in-game, but that's no biggy.The
calibration would remain 100% valid until the such a time as the cameras are
repositioned).
A face AVI could be recorded as part of the replay.
TODO: find web-cam; get hold of video stream
NOTE: be careful not to open a
second tracking app while the camera is already in use (eg. TrackIR while the
game is running); this breaks things.
1-Aug-2007
2
TrackIR arrived this
afternoon.Does its job
wonderfully.I just need a
head-mounted stereo display now and I'm all VR'd up! >:DSeriously, I'm going to start looking for
an HMD, because TrackIR's angular tracking range is really good (wasted on a
monitor).
The TrackClip fits my headphones perfectly too.It's all groovy.
Both say 26° diagonal FOV.
Don't even know if they can be fed dual inputs though.
TODO: playbacks:
- record head track data
- allow custom music track (wav / mp3) so player can record head movements
in time with the music (and just generally, record themselves playing to a
score as it were) ?
TODO: use the TrackIR camera
LEDs for something (eg. low ammo warning / homing device?)
Camera details:
355x290, 120fps
Installed the OptiTrack
SDK.
From what I can tell, there are no licensing restrictions to stop me using
it in the game.
Started porting the alarmingly
convenient Optitrack sample into CHeadTracker
(HeadTracker.cpp & h)
31-Jul-2007
3
Installed VS2005 SP1 in the hope
that it would fix my macros ... it did!YAY!
Had to load my macros project again after installing the patch.
Just noticed a complication in
the motion blur:
reflections, specular & shadows currently get blurred according to the
receiver's screen-space velocity, not the caster's.
Loads of improvements to
ground.fx, which is now getting towards looking magnificent.
- improved the way the ripple/impact normals are combined with the ground
normals (TODO: only lerp TS XY)
- impact normals are now coming through much more clearly, even on
non-puddly ground.That's what I want,
nice & busy.
- patchwork normals are coming through true now, which strengthens the POM
back to the way it should be, nice & rugged (flattened normals somehow
reduce the apparent depth, a lot)
- TODO?: cheapen dimple influence in
PSReadGroundTopology using the lerp TS XY method<-DONE
(9.9.7)
- TODO: fix refraction-vs-POM glitch now visible around road paint (I think
this would've always been there, just wasn't visible with the flattened patchwork normals)
As a result of the ground.fx
improvements, specular patches are bigger and more ragged (GOOD).This has a knock-on effect on the exposure
control though (more light in the frame)
TODO: compensate
It also shows up the edges of the lighting
geometry; TODO: can the mask texture be improved?
Had a look at some TrackIR /
Track-IR demo videos, looks like something I should start experimenting with:
http://youtube.com/watch?v=_AO0F5sLdVM
TODO: in the stereoscopic modes,
make use of the real-world head positioning & tilt data, RAR!
TODO?: aim using head tracking (crosshair moves to point on screen being
looked at)?
Couldn't resist - bought a
TrackIR 4 PRO and the head-mounted USB IR light thing which should fit nicely
onto my beloved HD-25s.Order info
etc:
FTP: Removed
ReleaseLog.txt.Will continue to use
local copy for drafting though.
Fixed moblur jittering by
generating the velocity image earlier.
TODO: fix the zbuffer corruption resulting from this
move<-fixed (31st): GenerateVelocityImage
currently has to be done before after CRainImpact::Prep to avoid corrupting
the ground reflection.Don't know why
yet.FIXME?
TODO: previous frame
persistence
TODO: get the light marker spheres included in the motion blur
TODO: deal with falling rain motion blur
TODO: try writing player mesh's velocity to MRT[3] so it can be cancelled
out from the final screen-space velocity
29-Jul-2007
11.6667
Spent the day adding motion blur
(based on camera velocities).Still
plenty to do, but it's looking cool and adds a sense of "urgency"
to the action; hard to explain.
I'm going for eye-style motion blur rather than film style, the difference
being that eye-style is continuous and requires that lights be able to trail
seamlessly over several frames (the trailing is done in HD range which will
allow lights to trail farther & brighter than everything else).
Film-style is much simpler, and it will be available as an option
(motionBlur\type).
Later on, enemies and other moving objects might write their velocities to
MRT[3] to be combined with the camera velocities.
Added:
MotionBlur.fx
Velocity.fx
MotionBlur.cpp & h (CMotionBlur, an instance of which belongs to the
environment)
TODO: use FRAME ort (only used
for deghosting) as MOTIONBLUR ort
DeferredLightingApply.fx: found
a mul (mist) that should be moved into the VS instead (will require some
rejigging/hardcoding)
MotionBlur.fx: When trying to
use a texture sample to control number of loop iterations, got this (even
with the sample saturated):
"c:\Noctambule\Source\HLSL\MotionBlur.fx(128,12): warning X3553: Can't
use gradient instructions in loops with break, forcing loop to unroll
c:\Noctambule\Source\HLSL\MotionBlur.fx(126,2): error X3511: Unable to
unroll loop, loop does not appear to terminate in a timely manner (1024
iterations)"
Fixed it by using a constant
number of iterations and breaking out when the required number have been done.
28-Jul-2007
7.33333
TODO: get VS2005 SP1 (432 MB)
and see if it fixes my macros, which still aren't working.<-(31st)
it did, YAY!
#revision 651
Ground water flow now working
well; just need to add kerbside streams.
Flow is now framerate-independent and pausable.
Tried making the wind ripples follow the fluctuating speed of the wind, but
it just looked downright wrong.It's
because it's a single rigid layer; when its velocity changes it draws
attention to the fact that it's scrolling uniformly rather than actually
rippling.
TODO?: procedurally animate the wind ripples texture?
Nice to see that PNG compression
does the sort of stuff I thought it did:
postCurve.png:
256*16 = 3.68 KB
256*1 = 3.57 KB
Added option HDR\curve
below: started playing about
with post-processing (currently just curves; independent R/G/B is
crucial).
Took me a while to get the hang of it but it's now giving really powerful
results.The main thing I'll use it
for is to increase the vibrancy of the scene.When I take the post-processing off, the game looks dull and muddy in
comparison.The lights feel really
dazzling now, as they should - and look how it brings out the rain.
PS - ignore the timers; the shot on the left is paused and the other one
isn't.
27-Jul-2007
3.25
UINT % [+ve int constant] - [+ve
int constant] = UINT !!!!
It's tricky, but the ground
water flow is now starting to look promising.
- per-pixel UV scrolling based on vert normals; speed is multiplied by
water depth
- to maintain texture cohesion, I'll use two short (< 1 second) cross-fading scroll periods
- there's also a wind ripple layer with uniform direction.This fades in to 100% influence where the
gravity flow has speed=0, to hide static ripples
TODO:
- get the cross-fading in<-done (28th)
- tie wind ripple velocity into wind system<-no
(28th)
- add kerbside streams (these will have their own velocities that override
the gravity flow)
26-Jul-2007
2.3333
Now that's interesting.I've been away on holiday for a month and
now my Visual Studio macros don't work.
I search the devlog and see that the last time this happened was also after
I'd been on holiday for a month.
Anyway, I finally did my 'field
trip' - yay!
It was mostly an experiment to see how I'd cope thrown into a
French-speaking environment on my own - whether it would stress / frighten me
or whatever.But it went really well
and I enjoyed it.
Would have liked more opportunities for conversation though.My most-used phrases (especially round
Pigalle) were "Non merci, NON merci" and "Désolé, je ne fume pas"
;)
#revision 647
Well, it's hacky, but I've added
a way of specifying wrap/mirror addressing per texture.
If the texture's name contains #Um then the effect instance sets U mirror addressing on stages 0
and 1, otherwise it sets wrapping on stages 0 & 1.Can't set the addressing on the correct
stage when each texture is 'set' because the effect instances store pointers
to D3D textures not CTextures (and that can't easily be changed).
This system could be extended later to include the likes of #Vm (mirror V),
#Uc (clamp U), and any other attributes I wanted to control per-texture
rather than per effect instance.
Added the # code info to TextureProductionRules.txt:
Eg., the colour map of the
'kerbs' texture is now renamed kerbs_#Um.tga (&.psd) (only one of the maps needs the # code).
Its production folder is still just called 'kerbs'.
Saw v12.0 running on M's new
nVidia SM4 card.No performance
worries at all on that system, which is great to see :)
When it's running smoothly, it's just crying out for motion blur....
That card can blend to 2TTT targets (unlike the SM3 nVidias), which is also
reassuring.
TODO: Get the ground water
flowing properly now.Paint velocities
into the dimple objects?
27-Jun-2007
4
Thought it would be a nice idea
to use the town aerial as a backdrop for the Max schematic view - so I know
roughly where to find things.
Unfortunately Max is using some ridiculous method for displaying the zoomed
image.It's so chuggy I can't use it!
>:(In fact it runs out of memory
if I zoom in and then try to pan around!MA-AX!!That would have been a really helpful
feature.
Max: When renaming nodes, don't
forget about the LANDSCAPE_FRAMENAME_ defines!!
TODO: get rid of those defines and use some new
properties on NoctambuleScenic instead?
WordPad: Crash on ctrl+a, ctrl+f
in leam.x
I think I need more RAM.LOTS MORE.
MAX: FX exporter converts '#' in
node names to '_'.So don't use hashes
in node names.
Got the world back into a
working state.
(below) Got my prototype kerb
texture applied.Needs a few
adjustments, but the patchwork system and its POM are both doing a fine job
:)The pic on the right is a reference
photo.I need more puddles!
I put a wee smidgeon of a curve on that grab to darken the bottom end.This seems to be a very effective way of
making game images look more real somehow.TODO: apply a similar curve in post processing!
:D <-done(28.7.7)
TODO: some way of specifying
wrap/mirror addressing per texture?Mirror is good for kerbstones (as used above)<-done,
26th of July
26-Jun-2007
2
Made some kerb textures.
TODO: get the John Street north
kerb decals lining up<-done
TODO: nice wall chart reminding
me how on earth that world file is built up! (it's a towering maze of
Booleans and modifiers)
25-Jun-2007
3.16667
#revision 643
Yay, fixed the memory leak on
reload.Just had to release each
enumerated CEffect before reloading it.
Noticed that the number of unfreed allocations wasn't multiplying with the
number of reloads I did.This turned
out to be an indication that it was a 'reference leak' - might be worth
remembering that for the future.
TODO: merge-away GroundPOM.fx<-done
Another possible enemy type:
SPIDERS.I just evicted one the size
of a horse!
Added option landscape\POM_LOD
Improved the wall POM LOD so you
can hardly see it now (although it's very much there and doing a good job).
Got rid of GroundPOM.fx (now
merged with Ground.fx using uniform variables)
Put the ground & wall
mipping back the way it should be
TODO: ground POM LOD
24-Jun-2007
8.25
Site: Spent far too long
(logged) overhauling the release log.My thinking is that it could be a good source of hits (mentions all
the game's features in un-abbreviated form), so it deserves to have nice
pictures and links.
Added subdomains http://releaselog.programmerart.org and http://download.programmerart.org, both pointing at the release log.They don't seem to have popped into action
just yet though....
Dev log now links to main page,
release log and mailing list.
Seems I'm running out of web
space?Had to delete some stuff.I thought I had a couple of gigs or so, but
I must've been wrong.
Had a nice idea that the phone
box on Windsor Street could be the player's first 'objective'.You could start the game clutching a
rain-soaked piece of paper with phone number on it, and when you dial that
number in the phone box you get a nice bit of voice-acting 'business' on the
other end telling you what's going on and where to go.
Been thinking about audio quite
a bit recently.Would be great to
somehow make use of the real location for recording....
TODO?: use key choice
(WASD/cursors) to guess the handedness of the player?Is there a way of asking Windows whether or
not the mouse buttons have been swapped?
Eventually got mesh-material
E&C working - eventually!(in proper words that means I can now edit
my scenery materials on the fly).
TODO: find that last little memory leak.Something to do with effect instance
buffers?
23-Jun-2007
11.6667
Wall POM is now lodded pretty
much the same as the original ATi source.Big speed-up!!
Changed the enumerated effects
to use CEffect (same as the meshes).This threw up unexpected device-object-handling hassles; eventually
worked out the problem:
-DXUTCreate3DEnvironment creates the device; then the
effects get created from this function
-OnResetDevice is called but no-one's home because g_game.pModeStack ==
NULL
-DXUTCreateDevice tries to reset the device; can't because effects haven't
been freed
Shifted effect creation from the device-created stage into the
restore-device-objects stage.
DX control panel: 'Break on D3D9
error' works and is reet handy!
Direct3D9: (ERROR) :ps_3_0
shader can only be used with vs_3_0+ shader (SWVP or HWVP), or transformed
vertices.
first frame,
CModeEnvironment::RenderEye:
Direct3D9: (ERROR) :Viewport outside the render target surface
Direct3D9: (ERROR) :Clear failed.
Lights can now only be toggled
on & off while they're visible in the main draw (ie. not behind a wall)
#revision
628
Removed a few zbuffer clears that weren't needed.The main Z buffer is now only cleared in the following places
(projectors use a private Z buffer) :
- CModeEnvironment::RenderEye (start)
- CModeEnvironment::RenderSubset (after generating reflection image; TODO:
private zbuffer for reflections)
TODO: rain impact frequency
should vary according to the downwards rain speed and rain density.Also try bending the rate of the audio
loops to tie in with this.
TODO: varying rain density (scale the cells?)
Decided to get E&C working
on the landscape materials, since the recent POM work has been slowed-down by
having to reload all the time.This
turned out to be fairly complicated because of all the nice cacheing of
effects, handles etc. that I stole from the DX StateManager sample.
Had to change a load of mesh & mesh & material code, but this gave
me a good chance to re-familiarise myself with it.It should be almost ready for the E&C
to work now...
TODO: I'll
probably spend a large part of tomorrow continuing my spring-clean of the
landscape code.I think there's a lot of stuff that can be merged together (there's
two mesh systems there for a start)
- CMesh and CMeshObject
- CMesh::pEffectInstanceBuffer and CMeshObject::bufEffectInstances !
- CMaterial and CEffectInstance
TODO: get E&C working on
the landscape materials.Get POM LOD option working.
TODO: if there's any time left
over, get cracking on those wall patchworks (this will probably begin with a
spring-clean of the patchwork code, but that's not too big or messy).
22-Jun-2007
2
Changed my mind; I don't want to
support different mapping scales as such.If I want to make a
texture bigger I'll increase the 'WS Graphic' dimensions on the
material.That provides the parallax
PS with the information it needs to offset each pixel's position output.Tested this on an enlargened version of the
'bumpyDefault' material.
So, to reiterate:
My UVW modifiers should use a
width/height/depth of PARALLAX_REFERENCE_TEXTURE_DIMENSION, which is set to
the default dimension for the UVW modifier, 1 metre.
For the benefit of shadows, eg
BC window pod, the POM offsets the pixels' positions OUTWARDS, otherwise a
gap would appear where the wall pixels' depths are <1.The rest of the POM effect works by
pretending the pixels are pushed inwards, but there's no benefit to matching
up with that here (the inwards thing is a weak point of POM and a source of
artefacts).
TODO: consider getting all these POM surfaces to use displacement mapping
on SM4.
POM now influences depth output
in WOWvx mode.
Shadows landing on the big
chunky test walls now join badly with the ground (as they should).Also the 3DTV depth joins badly for the
same reason.
I really want to work on
something meaty this weekend, rather than tweaking and polishing weather
effects.
I want to work on wall patchworks, which lead onto texture
patchworks (composite textures, like frozen
orthographic patchworks) and, my favourite, posters (text surfaces)!I'm
getting a bit sick of all these untextured buildings with their silly bumpy
placeholder walls.Getting the wall
patchwork system in place will let me start texturing buildings, at
last.
A great thing about these patchworks is that you can start texturing stuff
before the before the modelling's even done.In the case of the walls, the patchwork texturing will help me
mark-out the shapes that need to be modelled later.A really back-to-front way of working, but
I think this will be loads easier than the conventional way of doing things.
Or should I do the second pass
on the shadows?
TOFIX: POM changes tonight have
caused white depth flickers round the test spheres in WOWvx mode.Some kind of clamping needed?I think the same can be seen when very
close against a wall (ie. shallow angles)<-FIXED
21-Jun-2007
3.5
Unrelated to the code-signing
lark, I also noticed that you can delete any of the game's files (even the
exe) and it will put them back
when you next run the game from the installed shortcut!It repairs
itself!
Doesn't repair any alterations you've made to
files, just restores missing files.
It's something clever about the 'target' details of the shortcut.You can put a copy the shortcut anywhere
and it still works.Running the exe
directly doesn't do any of that.
Removed unused sewage cover
textures from graphics\final
Running the game from a file://
link, GetCurrentDirectoryA returns C:\Documents and
Settings\user\Desktop\
But that's not the location of the exe or the
web page containing the link (?)
If you run the exe from a bat file, GetCurrentDirectoryA returns the location of the bat
file.
Running the exe from a Windows
shortcut always works fine.
Out of curiosity, tried running
from a bat file but hardcoding the current directory to
C:\noctambule\source\debug.
This found gameroot.dat but crashed in CAnimUser::RestoreDeviceObjects()
because this->pMesh->pAnimControllerToClone == NULL
???
Well I don't see how I'm
supposed to know where the exe is sitting if GetCurrentDirectoryA doesn't
tell me this.But there's no point
wasting time this evening trying to work it out.
I've added a helpful fatal error message that's displayed if you try to run
the game from a web link or a bat file.
I'll ask around and see if anyone knows how to find the exe's location.
Shaders.sln - (project
settings\FX build rule) optimisations are now enabled even in debug, to
prevent the parallax glitch described on the 15th.Was getting today this when using ctrl-F7
rather than shaderbuild.exe (F5)
Decided to get the POM affecting
position output, since the flat shadows on bumpy walls had been bugging me
for a while.
Slightly more complicated than I'd imagined (simply offsetting each pixel
along the face normal according to height is wrong), but it's working now except that it doesn't yet take into account mapping
scale.And I do want to support non-unit
mapping scale because it increases texture mileage and works fine with the
POM.
Just need to decide how to communicate the mapping scale to the shaders....<-see 22nd
MAX: normals are always exported
at unit length
20-Jun-2007
3
Spent the whole evening trying
to fix the "Publisher: Unknown Publisher" nonsense on my download
dialogue.
Found out a bit about code signing using X.509 publisher
certificates.
Managed to generate a test certificate and sign my installer using it, so
I'll do that for future releases although you're not supposed to.This gives a slight improvement: the
'Unknown Publisher' is now a link to the certificate containing some of my
details (not sure how to prevent it saying "email: none"
though.
I'm also timestamping the signatures using this address:
http://timestamp.comodoca.com/authenticode
(below) Here's what I really
want though - is it so much to ask???As far as I can gather, this would require me to buy a publisher
certificate from a certification authority like VeriSign / Comodo / Thawte.And those certificates cost hundreds of
pounds :(I'm not even sure if an
individual can obtain a publisher certificate, although MS's AuthentiCode
docs say you can.
If anyone knows a cheaper way of achieving this result, please do get in
touch!
EXCEL: seems to set pictures to
150% size when you add them; scale to 75% to get them back
19-Jun-2007
0
Got a failing assert( anaglyphDeghostingPairIndex !=
EANAGLYPHDEGHOSTINGPAIR_NUM )
when running a network copy of v12.0 via a web link.Installed locally it worked fine, and I
think the network exe worked fine if you ran it directly.
Can't see anything wrong with the code...
Ah, right.The game just doesn't
support launching via a file:// link, simple as
that.Can't find any files if you do
this.
I wonder how I fix that?How do I
make the exe always know where it is?TODO: investigate
Quadro FX 550
On SO's Quadro FX 550, the game
crashed trying to create projector RTs.
Reducing numProjectors to 1 fixed this; the game then ran but apparently
using software rendering (1fps) (?)
Reducing various render target sizes got it all working properly.
TODO: include a 'compatibility' profile with his
settings or similar.TODO: auto-detection stuff......
I've now replaced the projector RT assert with a more informative fatal
error that points the player to Default.txt.
Other than that, I think v12.0
is behaving itself.Certainly works
just fine on my quadro.
18-Jun-2007
3.5
Removed
ReflectedDeferredLighting.fx (switched to using a uniform 'reflected'
parameter)
The FX file was actually already missing from Shaders.sln and SVN; a
'clean' made this apparent
Removed
DeferredLightingNoShadows.fx and ReflectedLandscape.fx
They weren't being used.
TODO: find out whether or not
1920x540 is possible, for WOWvx mode (only a small saving since I'm already
rendering to a 960*540 frame RT)
After cleaning the shader
project and removing the unused effects, the volume lighting bug became
consistent, happening from the debugger and in all configs.
CProjector::GenerateDepthImages
now restores the previous value of CModeEnvironment::s_pCurrentRenderData
before it returns. This fixes the volume lighting bug.
Now unsetting unused render
targets before generating volume lighting image.
(CLightManager::GenerateLighting)
Removed all the deleted FX files
from the installer.Removed the MAX_
stand-ins from the installer too
Spent far too long making a new
index screen.
#version
12.0
#revision 624
Released version 12.0.Will test this on the ol' Quadro tomorrow.
17-Jun-2007
12.1667
Tried setting up a
g_viewInfo.flags.flippedView, to replace the render-category tests in
RenderSubset, to control the backface culling mode.
For some reason it wasn't as straightforward as I'd thought.Will come back and fix this after the
build.
Volume lighting is now only
generated for draws that include mist or rain (was previously being generated
for reflections too).
CLightManager::GenerateLighting
In anaglyph mode:
- volume lighting is now desaturated as required, in
GenerateVolumetricLighting.
- the offscreen volume lighting image is now anaglyphic.Originally this was so that rain could use
both views of the volume lighting after the two environment draws, but (see
below) now it just gets rid of an RT clear on the second eye.
Ran into the problem that the
stereo rain, being drawn after the two scene views, had a z buffer that was
only correct for the second of the two views.
Fixed this by shifting rain into CModeEnvironment::RenderSubset where it
logically should be (and this went alarmingly smoothly).It simplifies & tidies things a lot,
which I'm pleased about.
The HDR down-scaled scene image used by the rain is now the anaglyphic one
from the end of the previous frame.This is one of the rare cases where a one-frame lag is absolutely fine
(I can't even spot it without displaying the RT).
Rain (including its lighting
& refraction) is now properly stereoscopic.
Started setting effect constants
at the 'apply options' stage, where appropriate, which makes a lot of
sense.
Added CRainFalling::ApplyOptions
#revision
622
Removed option stereoscopy\_3DTV_disparityPower
This was used in my initial quick mockup of the
WOWvx format; it doesn't have any place in the actual format.
Example WOWvx profile no longer
disables tone mapping (this was a bodge to prevent blackouts on nVidia cards,
which now seem to have stopped so I'm marking them as fixed in v12).
Added C3DTV (3DTV.cpp&h),
which now houses all the WOWvx code.
WOWvx mode now follows the
format specification in converting from view-space depth, to re-scaled
z-buffer depth, to 'disparity'.This
means that walls will no longer appear curved (especially when viewing the
screen from a side position).
Realised I could use my near and far effect distances as
the clip distances in the disparity formula.
I think last time, I did it more longwindedly, using the actual clip
distances and then converting out into the range I wanted - silly!
3DTV: There's now a single
placeholder header texture: header_custom.png.But I've left the
texture option in for now (stereoscopy\_3DTV_headerTexture), just in case it
could still be useful.E3DTVHEADERTEXTURE_CUSTOM is now the only valid value.
TODO: generate the header texture using player options.Adaptive focal point would be real good
(focus on the enemy closest to the crosshair if aiming; focus on crosshair
collision point if facing a wall, etc).
3DTV: header is now concealed as
it should be (drawn with colorwrite=blue)
Current format for the header texture is PNG with header bits in the blue
channel and masking in the red channel.No transparency.
Above: an example of the current
depth data in WOWvx mode (ESTEREOSCOPYMODE_3DTV).
(NB: this is not the aspect ratio or resolution used when driving the
display.)
Note the characters (right foot) and car connecting properly with the
ground.
This grab has not been run through any curves in Photoshop.The banding is an artefact added by Excel.
Argh, version 12.0 is halted at
the last minute by a graphical corruption bug that's sprung out of nowhere.
The volume lighting image is static junk unless the game is run from the
debugger :(
Grr.Hopefully v12 will go out
tomorrow night then.Much as I'd like
to start debugging the problem now, it's getting just a wee bit late...<-FIXED (18th)
16-Jun-2007
8.58333
Added option shadows\groundCastsShadows
(default = true)
This should probably be considered a 'luxury'
option for now.Need to time all those
redraws of the ground.
Ideally I'd use low-poly replacement shadow meshes (just the vertical kerb
faces in fact).Need to find a way of
generating these meshes automatically.
Having said all that, these ground shadows don't seem to be denting my
framerate at all.
#revision
613
Removed
PROJECTOR_DEBUG_DEPTH_IMAGES.It's
served its purpose and it doesn't work now anyway.
TODO: private Z buffer for
reflections (see Projector.cpp)Saves
clearing the main (larger) one.
Fixed a bug that was causing
ground patchwork decals to draw into projector depth images.
Removed MESHRENDERFLAG_SIBLINGS from MESHRENDERFLAG_DEFAULTS to prevent
similar bugs in future.
It says that quadros can do 16*
anisotropic filtering.So why was this
failing (with warning messages) on PC558?
Quadro listed in cardcaps.xls is a 4400....
TODO: get a dxdiag and caps
report from PC558.
Patchwork.fx: Set the mipping
& filtering back to anisotropic for now.Need to see whether or not my quadro reports
D3DPRASTERCAPS_ANISOTROPY, in order to know what's the best thing to do here.
Deghosting is currently broken; checked that this wasn't caused by the colorwrite changes in
zombie.fx / vertexcolour.fx.
Ordinary anaglyph is working ok though.
Light marker sprites now appear
in both stereo eyes.
(I had forgotten the 'g_game.cameraEye' parameter on
'pCamera->ApplyView' at the end of CModeEnvironment::RenderSubset.)
To get light marker sprites
correctly colour-filtered in anaglyph mode, moved the colorwrite restore from
ApplyDeferredLighting to RenderSubset
HLSL: good habit to qualify
uniform parameters with 'const'.
I'm also naming them 'uniform_...' to make it obvious that testing them is free.
Need to remember to test the uniform parameter
and not the effect constant itself !
#revision
616
The reason the mist wasn't
appearing in anaglyph mode was that the mist constants were being set on
DeferredLightingApply.fx, not AnaglyphDeferredLightingApply.fx.
Fixed this by converting DeferredLightingApply to use uniform parameters
controlling the effect version (normal,anaglyph,reflected) - instead of
multiple fx files with different defines.
This allowed me to get rid of AnaglyphDeferredLightingApply and
ReflectedDeferredLightingApply.
TODO: do the same with the rest of the FX file variants?It's much neater, and the cost of switching
the effect version is totally negligable.
Collision.cpp: added IntersectLineCircle
#revision
617
Made the mist properly
stereoscopic (CMist::Generate).It's
great!It strengthens the illusion of
3D by adding three layers of moving depth information.Hopefully the rain will strengthen it
further.
Tested stereo mist quite thoroughly.
Light marker sprites now honour
the 'sterescopy\saturation' option (was this the case in v11.1?).
CLightManager::DrawPlaceholderLightObjects
Fixed zombie sphere-frustum
culling (wasn't using correct centre point).
Also removed CZombie::visible.It
wasn't being used and it was misleading (it was set on every TestVisibility,
even when drawing to projector depth images).
Added '#enemies in main draw' debug display; checked that the zombie sphere
culling is correct.
TODO: zombie cullplane occlusion!!
TODO: stereo falling rain <-DONE (17th)
15-Jun-2007
2.75
#revision
609
Trying to copy the whole
Noctambule folder back from a DVD made at the office wasn't a great
idea.Looked like it was taking
forever to just to get through all the pngs in the rain streak
database!
I need to just transfer the folders I've changed.
Transferred Source, Profiles, and Docs
Chcked that all the adjusted
code I brought back from the office still works on my April 2006 DX SDK
Installed the June 2007 DX
SDK.Nothing new in the samples
really.
Had a quick look at the
open-source web tools at http://www.drupal.org/
TODO: Give them a try when I come to overhaul the site?
TODO: Also need to try Flash, which is likely to be a good thing to
use.
TODO: Find out about making web pages using Photoshop
Got rid of
ESHADEREFFECT_GROUNDREFLECTION, GroundReflection.fx, totally OOD
Got rid of
ESHADEREFFECT_PATCHWORKDIMPLEHEIGHTCOMBINE
TODO: in the replays, do that
shot where you just see the monster's grotesque shadow creeping along a wall,
you don't see the monster itself.
Also do the 'monster sneakup cam' bwahahaha!It's going to be so much fun coding the
'cinematography' for these replays.
The replay system will be very useful for making gameplay videos too, and
for benchmarking / optimising.
Well, with the June 2007 DX
SDK, with optimisations disabled on the fx compilation ( /Od ), my POM is
broken (ground and walls in pretty much the same
way).I thought normally things broke
when you switched optimisations ON?!
Doesn't matter a hoot though; I've now removed the switch in
ShaderBuild.cpp.
If I'd been outputting and source-controlling the disassembly text like I
used to, I'd be able to see what the difference was...
TODO for version 12 (to be
released this weekend) :
- deal with mipping/filtering (anisotropic) stuff
- deal with colorwrite (MRT) stuff
- ensure anaglyph modes still work (colorwrite)
- fix WOWvx mode (new header, Z-to-depth,
depth-to-disparity)
- scratch head over HDR\finalMultiplier (quadro) discrepancy.Need to check a 7900 or similar.Should have tested floating-point lighting
RTs on the quadro but forgot.
THEN, get the basic replays
working so I can record some "rain tech" videos ??Like all the cool kids are doing? ;P
Also get the ground water velocities working (more urgent I suppose).
14-Jun-2007
2.66667
quadro,
PC558,
April 2007 DX
Had a dream that I was Captain
Picard.Shouted at an office full of
people, then talked to Counsellor Troi.Woke up feeling confident and in control.Need to do that more often.
Severe rain this evening, so I
put the camera on charge and bought some cling-film.But as usual the rain stopped before it got
dark :(
Had a good look at it though, and I definitely feel I'm on the right track
with my rain effects.
- Streaks and impact splashes and were pretty much like mine.Impact splashes were a bit smaller, maybe a
bit more densely packed, and had a better draw distance ;)As the water got deeper, there was more in
the way of vertical spikes maybe 3-4" tall.Maybe you wouldn't see these if it was
windy.
- The flowing across all the ground looked spookily like my current rough
version.
- Very non-binary puddles, same as I'm going for now.
- Lots of water-in-the-eyes.Drips
gather on the eyebrows & eyelashes (big blurring towards top of view
especially) and drip off now & again.
There wasn't any real wind to look at unfortunately.
Fixed the reflection glitches (freezing up/cutting-out, see yesterday) :
In CAnimMesh::DrawMeshContainer, the block at the end that draws
non-skinned sub-objects was somehow causing it.This would have been used for the zombies'
eyeballs, which have not shown-up in game since like version 3 or
something.
Checked that it wasn't just the SetTransform(D3DTS_WORLD, ...
W00T! The fix above also fixes
the bad zombie normals!! ^_^I think that's actually the last of the
quadro bugs!
shifted a zbuffer clear in
CModeEnvironment::RenderSubset.Shifted it to just after GenerateReflectionImage.
IS IT NEEDED?
TOFIX: suspicious stutter every
couple of seconds
Version 12 is very close now
:)...Thank goodness; v11 looks really
out-of-date now, without these weather effects.
Checked that a Master build
works ok.
It started raining again after
midnight, so I got some photos & videos around Bedford Court.
Cling-film worked ok for waterproofing, but got very blurred by the
water.Maybe try the umbrella next
time:
Spent two hours (not logged) investigating a possible shader effect that
would displace surface detail in such a way as to follow the curves implied
by smooth vert normals.
Couldn't see a good way of getting the required information into the PS
(face normal, UV offset from face centre) without breaking the
stripping.
Other than that it looked like it would work.
Fixed the bulk of the shadow
junk by setting the backface culling mode depending on
whether or not the view is flipped.TODO: g_viewInfo needs to precalc and store this.
W00T! As a result of the fix
above, managed to get PROJECTORSHADOW_DEPTH_BIAS down to 0
!!!
Projector.h:changed PROJECTOR_ORT_FORMAT to a float
format because the Quadro somehow didn't have enough depth precision for the
shadows otherwise.16Fs work fine.NOTE: only red channel required.
Tried R32F but the quadro couldn't do it, wanted to switch to reference.
TOFIX: looked as if road lines
etc were rendering into projector images?!!!! <- FIXED
(16th)
ZBUFFER JUNK:
- Only happens when characters are allowed to draw.
- Only appears on characters, cars and the sky
- happens even if zbuffer is cleared before drawing the thing (eg.
player)
- happens even if zombie.fx and vertexcolour.fx use ztest ALWAYS
NOTE: no E&C on vertexcolour.fx
**** SORTED! ****
had to set
COLORWRITEENABLE = RED | GREEN | BLUE | ***ALPHA***;
in zombie.fx.Otherwise the
mistmatch causes a silent MRT failure on quadro.
Previously COLORWRITEENABLE wasn't getting set at all.
Made the same change in vertexcolour.fx for safety (otherwise I think it's
relying on draw order).
***TODO***: make this change more global and anaglyph-friendly (currently I
think it'll break anaglyph)
TOFIX: reflections are flicking
on and off on quadro, also occasionally freezing.doesn't seem to have been caused by the fix
above.
- happens even when paused (?!)
- happens with or without projectors
- only happens when characters are allowed to draw to the reflections
- remming-out the DrawFrame call in CAnimMesh::Render prevents it
<-FIXED (14th)
TOFIX: zombie normals are
coming out a bit wrong on quadro<-FIXED (14th)
Had to set HDR finalMultiplier
to 1.f (rather than 2.f) on quadro to get it to look right(?)Looks exactly right at 1.
12-Jun-2007
4.666
6.30->
quadro,
PC558,
April DX SDK
Had to rename any shader
functions called 'PixelShader' or 'VertexShader'.
Fixed the following error by
using the [unroll] attribute:
C:\Noctambule\Source\HLSL\RainLighting.fx(109,13):
error X6077: texld/texldb/texldp/dsx/dsy instructions with r# as source
cannot be used inside dynamic conditional 'if' blocks, dynamic conditional
subroutine calls, or loop/rep with break …
... "These issues will be
fixed in a future release of the SDK.
If you have a loop in your pixel shader, and the compiler generates error
X6077 where it previously did not generate this error, try one of the
following work arounds:
- Change the loop to the form for(int=0;i<cnt;i++)
- If possible, unroll the loop using the [unroll] attribute."
Noticed some bizarre white
pixels on the rain lighting image.Fixed these by adding a 'max' to the returned RGB of
DeferredLightingApply.fx (rain splashes).A negative colour was being output because of the bit where the
pixel's diffuse lighting colour is normalised (normalising black causes a
db0).
Fixed the following errors, by
changing anisotropic filters to linear.
Changed mips to NONE though; linear mipping on the road lines comes out
bad.
Debug D3D, Patchwork.fx, FindNextValidTechnique:
NOTE: I'VE PUT THIS BACK TO ANISOTROPIC FOR NOW.NOT SURE WHAT THE BEST THING TO DO IS.DEPENDS WHETHER OR NOT MY QUADRO REPORTS
D3DPRASTERCAPS_ANISOTROPY....
… Direct3D9: (ERROR)
:Unsupported mag filter.
Direct3D9: (ERROR) :Invalid texture sampler state value. SetSamplerState
failed.
Direct3D9: (ERROR) :SetSamplerState failed.
First-chance exception at 0x7c812a5b in Noctambule.exe: Microsoft C++
exception: long at memory location 0x0012c9c8..
(etc)
TODO: fix this?:
"Direct3D9: (WARN) :Stream 0 stride and vertex size, computed from the
current vertex declaration or FVF, are different, which might not work with
pre-DX8 drivers"
Fixed the following, by changing
SFallbackShadowVert.TODO: test the
'lightPos' member now (see shader) :
"Direct3D9: (ERROR) :Vertex shader function usage
(D3DDECLUSAGE_TEXCOORD,0) does not have corresponding usage in the current
vertex declaration"
(search for remmed-out 'V's on
'DrawSubset')
TODO: fix this:
"Direct3D9: (ERROR) :Vertex shader function usage
(D3DDECLUSAGE_COLOR,0) does not have corresponding usage in the current
vertex declaration"
zbuffery wierdness only happens
when characters are allowed to draw.
Only appears on characters, cars and the sky
TODO: why are the sewage covers
not glinting properly on Quadro?Because I turned mipping off?
Shadow junk is a backface
culling issue.No idea why it should
be different on different machines.But in theory I need to work out whether the view matrix is flipped or
not, for each frustum face, to know which backface culling mode to use.
Maybe there's a difference in
zbuffer precision (actually it does look that way), and it's working on my
card by fluke?The 'big white light'
does show problems on my card....
11-Jun-2007
3
Spent far too long getting a
background image onto the front page of the site (+ lecauchemar.com)
Had to increase the 'target schema' to IE 5 for this (was on 3)
Got rid of some old data from
the devlog file properties, which were showing up in the HTML: company name.
Improved the rain textures, and
I'm more-or-less happy with them now.It's glinting better and it has a nice motion blur.Also tweaked the shaders; removed a couple
of bodges now the streak texture is better.
Need to stop fine-tweaking now, it's taking up too much time.
At the moment, I'm averaging
about 50/60K rain triangles per frame (7 cells).I'll put in timers before too long...
Downloaded a database of rain
streak photos from cs.columbia.edu.Interesting to see, but not directly useful for me as textures like I
was hoping.
TODO: try breaking-up that mist
texture
TODO: improve wind
TODO: is there some kind of lengthening I can do on the
fallback shadows...?Would need to
take into account distance from the caster (no stretching at the point where
the 'shadow' meets the ground plane)....
10-Jun-2007
10.166
Decided to get the rain
celled-up & moving before coming back to finish the lighting.
Began
by thinking about how to sync the falling raindrops with the rain impact
splashes, since this could affect how I cell-up and move the rain.A few possible approaches:
1. Basically make the falling
rain match the existing impact effects.Arrange the raindrops into columns with an XZ area of 300*300 to match
the impact tile.Impacts are read from
a LUT matching the falling raindrop positions.
Disadvantages:
- Rain cells would be smaller than I'd like; might be able to spot spatial
repetition.
- Impacts would be temporally repeating
- Impacts would only match at the ground plane!
- Forces impact splashes to slide along the ground to match the rain
movement; don't know exactly how this would look yet
2. Completely new rain impact
effect made by redrawing the falling rain triangles with different
shaders.Pixel shader tests the
position buffer to form a splash as the triangle hits the scenery.
Disadvantages:
- I'd be starting from scratch again
- I'm reluctant to draw all those rain triangles again (or most of them;
maybe upper cells could be skipped)
- How's each splash image going to be formed to
account for view angle?That was hard
enough the first time.
- probably more expensive than the existing splash effect
3. Use the existing impact
effect, but replace the planar-mapped impact dots with 'proximity spots' on
falling rain triangles.The pixel
shader tests the position buffer to generate a circular dot where the
triangle is close to the scenery.This
then goes through all the same extrusion, lighting etc that I'm already
using, to form a splash.
Disadvantages:
- Argh, the impact ripples won't match the
impact splashes :(That's no use.
...Nah, beats me.I can't see an affordable way of making
these raindrops match with their impact effects.
Doesn't actually bother me though; I don't think it'd be very noticeable
anyway.And I certainly wouldn't want
to compromise the quality of the falling rain, or the impacts, to get the two
to match.
Got the rain celled-up &
moving.Looks great, and really
completes the scene.
The fact that the drops don't match-up with their impact effects is
pretty-much un-noticeable as expected.
Gave the rain a touch of
constant grey which really helps it in the darker areas (same as I did with
the mist).
NOTE: In an array of VEC3, the
elements are not 16-byte aligned.The
spacing is 12 bytes which is sizeof(VEC3)
Arranged the rain cells into a
spheroid formation.
Now drawing back to front.
Now culling cells below the ground effect plane.
Streak texture now scrolls properly.
Refraction now decreases with distance from viewpoint.
Added RainLighting.fx which
forms the rain lighting image by blurring an HDR scene image and adding the
volume lighting, onto a tiny 2TTT render target.
This lighting image works well.Also
removes a texture sample from RainFalling.fx (background and volume lighting
are now combined into this new image).
Improved the glintiness of the
rain by brightening the refraction according to the Y normal.
TODO: Improve rain streak
texture a bit.Smaller/blurrier
alpha?Motion-blur the alpha?Less crinkle-cut.<-done(11th)
9-Jun-2007
9.833
HLSL: sampler_state: the
brackets around the 'Texture' property value are manditory.Why is that?
Got some practice at using paths
in Photoshop.
(below) Got the rain textured
and lit in a rough form.Still trying
to get the lighting right, not there yet.There's a distinctive glinting I want to see when the drops pass close
to the lights.
Refraction is offset down on the Y slightly to reinforce the rain impact
splashes (eg. looks good on the cars).
TODO: separate rain-lighting image (MRT this in with
the mist lighting to prevent duplicated shader work?)<-done(10th, RainLighting.fx)
TODO: give rain impact splashes the same refraction as the falling
rain?
TODO: take into account player's angular view velocity when stretching the
rain
8-Jun-2007
3.25
Got the rain triangles properly
shaped & angled by the vertex shader.
7-Jun-2007
4
Added CRainFalling
(RainFalling.cpp&h), the main falling rain effect.I'm reserving the name 'Rain.cpp&h' in
case I need a rain manager or general rain code of some sort later.Besides, CRain would be a bit of a vague
name for any of the rain effects, even the main one.
Used a new neat format for
RainFalling.fx, which I'll use for new HLSL effects from now on.Till now, the effect files have been a wee
bit messy, but I expect I'll do some tidying when I do my
compatibility+optimisation pass (hopefully soon).
option weather
\ mist now works properly.
Added option weather
\ rainFalling.
Got the rain geometry drawing,
currently just a bunch of simple triangles (lots, roughly a zillion).
This is the first effect of mine that uses a D3D vertex buffer.
6-Jun-2007
4
Decided not to persevere with
the current flukey ground spray effect.I'll hopefully come back later and do a proper one instead.
Also decided not to get too tied
up with wind / mist turbulence just yet.It's a bit robotic-looking at the moment but I'll get a better idea
what it needs once some other effects have gone in.
So then - next weather
effect.And it's probably the most
important one: falling rain.
Planned the falling rain.Looked at a lot of reference.
I'm keen to avoid the familiar imperfections of the old-fashioned rain
techniques (ie. I don't want to be followed around by some kind of 'scrolling
lamp shade').
The current plan, stupid as it sounds, is to pour a zillion little
triangles through the view frustum.These will be batteried into huge triangle lists that don't require
any update from the CPU.All the
wizardry of shaping, angling and texturing the triangles is done in the
shaders.
This approach has a bunch of advantages; one of them
is that I should be able to sync the falling raindrops with their own impact
splashes & ripples !!!!<-(or not,
10th)It will also play very nicely
with stereoscopy, and bullet-time if I have any.
Got a plan for the texturing too that should look just right....
5-Jun-2007
2.75
Fixed a very slight (but
annoying) brightness glitch that could be seen each time a mist sphere
wrapped around on the Z.Just had to
increase the minimum sphere scale a bit.
Fixed the crazy polys over the
screen in Master (silly mistake in GenerateVolumetricLighting).
Had a go at lighting the
ground-level spray.Bit more work
needed here; not convinced yet that it'll be usable.
The waves of spray are a fluke effect, and I expect something deliberate
will work better instead.
TODO: try a second pass in VolumeLighting which lights
the spray with a single colour (no influence from the lights).
TODO: consider diffusing
specular highlights (or something) into the mist lighting image.
Got the Z movement of the mist
all working properly, tied into the wind and player movement.The whole thing looks pretty convincing
now; I'm reet pleased with it.As well
as being really valuable in this game, it's a handy effect to have in storage
for the future.
Had to go up to 3 cubemap samples per pixel; 2 didn't quite cut it.
At ground level, there is now a pretty good rendition of those waves of
spray you get during high winds and heavy rain (almost like sea spray).TODO: try giving these some kind of
lighting? (currently they're invisible because there's no mist lighting at
ground level)
Tidied render target setup (some
were getting set up in Main.cpp rather than Options.cpp).
Got weird nonsense polys over
the screen in Master.This is
new.TODO: FIX THIS!!!<-FIXED(5th) (silly mistake in GenerateVolumetricLighting)
above: rain mist.Not an effect that particularly likes
having its picture taken; you have to see it moving.The banding is an artefact added by Excel -
it's nice & smooth in-game.
Mist lighting now uses
2-10-10-10 if available.Otherwise, it
uses float16s if weather\mistLightingRTFavourFloat, else 8888 (which I
already know is usable for this effect).
3-Jun-2007
5.25
SVN: Used 'update to revision'
on the Noctambule folder and found it caused Noctambule.sln to lose its
project.This has happened before I
think.
<-need to close visual studio when updating to an old revision
Fixed the blockiness of the mist
silhouettes, by reluctantly shifting the mist generation into
DeferredLightingApply.
There is still a teeny bit of blockiness, from the volume lighting, but I
can certainly live with it.Anyone
with frame time to spare can remove this remaining blockiness by setting weather\mistLightingRT to [/1].
Lights can now be toggled on
& off by shooting at them.This is
stand-in behaviour till you can shoot them out properly with weapons.
TODO: Exposure control becomes a
lot more relevant now that the lights can be switched off.Need a fix for the nVidia cards in v12.
Devised a means of blowing the
mist into and away from the viewpoint (as well as just across).Looks fantastic.
TODO: hook this up to the wind-minus-camera
velocity.Add a numMistLayers option?<-done(4th)
2-Jun-2007
7
Got the latest ATI Catalyst
control stuff installed; it had sorta gone missing.v11.1 is quite silky with the overclocking
turned on.
Downloaded ATI GPU analyzer
thing and CubeMapGen.
CubeMapGen: Z- is FR, Z+ is BK
Set up a Tools\CubeMapGen
folder, containing the CubeMapGen example project which I'll extend to do
whatever I need from it.
Tried using the ordinary
CubeMapGen.exe in command-line mode.Couldn't get it to import cube face textures; tried all sorts.
VolumeLighting.fx: now using
squared falloff
Above: a lit mist image before
it's blended into the scene.Looks
nice when it's blowing around.The two
parallax layers give a good impression of volume.Texture is a bit solid in this example;
TODO: try breaking it up using vertex colour on the cube map sphere in Max.
MAX: found out about reflect/refract maps, which render
cube map faces for you - cool!Saves
me getting my hands all covered in MaxScript.
Got a beautiful cube map result
from Max's 3D smoke map after playing with the parameters a bit.Inverted smoke colour works really well for
this mist, similar to how I was using Photoshop minimum filter before.Makes it wispy.
Started dealing with the
blockiness of the mist silhouettes....
TODO: maybe try directional
lighting on the mist.Copy unlit mist
image to a small RT and generate normals from that.<-nah,
looks fine without it, and I've got better things to spend my frame time on
1-Jun-2007
2.75
Got the volume lighting working
on the mist, very nice.Even the
volume lighting image on its own makes a decent fog effect (below).
It definitely pays to think of fog/mist as a way of lighting
the atmosphere, rather than a cloud of grey muck
in the background.
above: volume lighting image
used by the mist, will also be used to light the rain and drips.
Noticed some accidental colour
dispersion in the mist around some lights
TODO: should be easy to add deliberate dispersion in VolumeLighting.fx
TODO:
- tweak mist lighting/blending<-DONE
- deal with ugly colour clamping in the mist
lighting<-DONE (by moving mist stuff into
DeferredLightingApply)
- fix blocky mist silhouettes? (might be too
expensive to fix this)<-DONE (by moving mist
stuff into DeferredLightingApply)
- fix mist texture(s), even if still just a
placeholder<-DONE (by moving mist stuff into
DeferredLightingApply)
- improve wind.Wants to be more
blustery.Maybe something needs to be
done in the VS/PS to make it more swirly, less cohesive.Need to think about how the blustering rain
is going to tie-in to the mist wind (maybe wait till the rain has gone in
before tweaking the wind?)
30-May-2007
Went (well, crossed
the road) to see 28 Weeks Later.It was research!
Mmmm, that music would be great to use in a teaser video, I can see it
now....
29-May-2007
0.5
Busy sorting holiday-related
stuff.
Got another one of those
lockup-reset crashes, similar conditions as before, game running in debug
(windowed), behind maximised windows:
"Error Message: STOP 0x000000EA THREAD_STUCK_IN_DEVICE_DRIVER
(Q293078)"
Tried to get latest Windows updates but this crashed explorer.exe.
28-May-2007
10
Added option weather\wind (true).
Added CWind (Wind.cpp&h),
the wind controller.I'm going to keep
this nice and simple.Currently just updates a single velocity
vector, and I think that's all I should need.
Added
CModeEnvironment::timeRunning, which should be used as the timer for all
environment features.
This pauses when the game is paused.It will reset to zero at the start of the replay.
TODO: Use the mist (recoloured)
as a dream effect for the intro.This
can fade seamlessly into the game.
Finally worked out why I have so
much trouble with vertex formats:
VEC3 is 16-byte aligned.
Add 4 bytes of explicit padding after any VEC3 in a vertex structure (or
better, pack something useful into the gap).
Use D3DFVF_XYZW
with a VEC3.
Use D3DFVF_TEXCOORDSIZE4 with a VEC3 (NOTE: still couldn't get this to work)
Never use D3DFVF_NORMAL with a VEC3.
Never remove the alignment from VEC3.
Got the mist moving fairly
nicely, more to do.
Mist lighting going well
TODO: fix up vertex formats, finish mist lighting
27-May-2007
6.75
Added options weather\mist (true) and weather\mistLightingRT (small).
Moved the PC into a quieter
room, which is loads better.Can't
hear my neighbour any more, yelling out of his window at passers-by, like a
dog barking at other dogs.That's the
reason I got no work done yesterday - couldn't concentrate, and ended up in a
terrible mood because of it.
Realised that the HDR system
doesn't contain the data I need to light my mist (+ rain etc).The luminance images are mono and I
couldn't get anything but constant solid colour out of them anyway; dunno
what I was doing wrong there.
So instead I'm going to form a small volumetric lighting image
(ERENDERTARGET_MISTLIGHTING), and probably mix the HDR bright pass onto
it.Not going to think about shadows
just yet...
Added CMist (Mist.cpp&h)
which controls the mist effects.
Devlog: Stopped jotting-down
French vocabulary.It was a useful
place to put it, but with the devlog being public and all...dunno, looked a
bit like weird haikus or something.
Added CNoise
(Noise.cpp&h).Ported from ATI SDK
(perlin.cpp)
Got static mist in, already
looking nice & eerie. Currently using a 360x180 panorama
texture.A cube map would be quicker
to look-up...TODO: see if a cube map would be worth the hassle (and seams).If not, use that ol' vertical spherize
(Photoshop) trick to remove an atan from GetPanoramaUV.
TODO: pack a different layer of
mist detail into each of the texture's colour components<-not
needed now, with the zooming system
TODO: get that mist a-swirling
26-May-2007
4.5
More little improvements to rain
splash lighting.The normals now
rotate according to the XZ view direction.
#revision 550
about to start on fog/mist.
MAX: Grr, 3D maps display as 2D
maps in the viewports.Have to render
to see them properly.
Is this a problem caused by the DX SDK trying to be helpful, or does Max
just have no way of drawing them properly in real time?
Tried Max 3D maps (on a sphere)
to see if they'd give me good smoke cubemaps.Couldn't get anything inspiring out of them.
Read up on some stuff on the Internet.Started work on the mist.
25-May-2007
4
Improved rain-splash
lighting.Made the lighting and the
normal-generation much cheaper.
Remembered that
deferredLighting.fx is THE most speed-critical pixel shader in the game.Simplifying its rain-splash lighting gave a
noticeable speedup.
Accidentally fixed the
fluffiness of the rain impact splashes :)They now look like proper rain spatters rather than lumps of cotton
wool.I'm no longer embarrassed to see
them.
The gist is that their brightness is now weighted by the 0-to-1-packed z
component of the splash normals (complete accident).This gives the droplets that important
bright outline effect, and it also hollows-out the shape of the splash.
Experimented with a refractive
version of the rain impact splashes, but it didn't look like it was taking me
anywhere.
TODO: Looks like the fallback
shadows are a frame behind?Something
definitely is, stand at the start point and look around, look at the ground
highlights.
24-May-2007
1.83333
Oops, v11.0 looks for fxo files
in C:\noctambule\....That's not going
to work.
It was UtilCreateEffectFromFile.Must've been a temp test I put in when shader E&C started playing
up?
MUST ALWAYS TEST BUILDS USING A RENAMED C:\NOCTAMBULE
FOLDER!!!
v11 also had a few missing fxos
and some weird duplicate texture entries, now fixed.
TODO: Find out why the
lightsource sprites sometimes don't appear.<-hopefully
that was the paths bug in v11.0
#version 11.1
Released version 11.1.Practically the same as v11.0 but without
the startup crash.
23-May-2007
0
got a blog at blogspot.com.Just experimenting with it to see if it
could be used for the devlog.
started a free trial of the
non-public extremetracking.com tracker.about £36 per year
22-May-2007
3.25
Turns out the rain deadline in a
week or so isn't THE rain deadline, which will be around October.Just as well really!
This one's just a sorta progress check.
Got a weird sudden black-screen
lockup while the game was in a window, behind a maximised Visual Studio
window.
Media Player was playing chunky wavs, if that's relevant.It went into a buzzing sound.
Rain impact splashes are now
plotted in the correct positions, so they now match up with the impact
ripples.
Previously most of the impact dots were going off-viewport.This is because I had been pretty sure the
viewport width was always the same as the display width regardless of RT
dimensions.Seems now this isn't the
case; the viewport dimensions in GenerateImpactImage are the RT dimensions.
Halved the number of particles because it was very dense after this fix.
Puddle ripples no longer move
with the viewpoint.
Reflection lag: Fly cam only; I
think it's the the ground plane
data that are 1 frame behind.
Decided to leave the existing
rain effects in their current rough state while I add the rest of them.
I reckoned it's better to have a full set in place so that I can get a feel
for how they'll play together, which ones are most important, etc, making it
easier to tweak and budget them.It'll
also look more impressive at the rain deadline to have a full set of rough
effects than to have a partial set of polished ones; I'm sure of that (it's
just like a painting).
So then, next
weather effect:
Start at the back.FOG / MIST.This'll add a huge dollop of atmosphere
& motion (includes wind).Strongly
tied into lighting too, as mentioned in the brief for the weather effects.It's a good'un.
If it weren't for the deadline in a week or so, I'd probably do swaying
streetlights first, but they're not as impressive to look at.
FOG / MIST: Remember:
- wind
- parallax
- directionally lit
- diffuses the bright pass
- etc
Site: Added the hit-tracker back
(same login etc).I think the
script/ActiveX security message only appears if you view a copy of the page
on the local machine.
TODO: Find out why the
lightsource sprites sometimes don't appear.<-hopefully
that was the paths bug in v11.0
20-May-2007
6.33333
Lights now inherit from
CNameUser and read their names in from the nwf.
IMPORTANT: CNameUser only holds a name string in DEBUG.
CAUTION: when flipping a
CullPlane, be careful not to flip any child objects with it!
Lights no longer use CullPlane
exclusions.It was causing some
mis-culling of the BT lights (when viewing from behind the west BT
wall).The exclusion system is really
just designed for buildings anyway.
CullPlanes: Removed upper plane
bounds check (so the planes now extend upwards to infinity as well as
downwards).Might not be permanent,
but it's great for testing cullplanes from a birds-eye view.It also helps to cull the tops of
streetlights.
TODO: reshape streetlight bounds to better match their (TODO) projection
mask.
Tweaked lighting again.Found BRIGHT_PASS_THRESHOLD
(HDRLighting.fx).
Disabled tone-mapping for now, because it's broken on
nVidia cards.
Added lights on Windsor Street
and Windsor Place.So in version 11
there's a figure-of-eight to explore.
Found that the north John Street
test kerb paint had lost its 90° rotation round the spline (???)Now fixed.
TODO: put one of those fancy
drag-around maps of the town on the site (see Google Maps)
CullPlanes referenced by meshes
& lights now work in non-debug builds (previously they were only holding
a name in debug builds!)
#version 11.0
Released version 11.Plenty of things wrong with it but it's
more impressive than the last build.
Stopped supporting the
English-language installer.Too much
hassle for no real benefit.
19-May-2007
7
Fixed the log converter (doubled
the size of its text buffer, now 2 meg).
Decided not to make the
buildings children of ground chunks.It would make it awkward to change the chunk size or to use ground
sub-objects.
#revision 538
C:\3dsmax6 is now under version
control.Can't risk losing/breaking
all those little scripts, plugins, settings files, etc.
Max: PointLight: now has a
CullPlane property like NoctambuleScenic.
Got this working in-game.
#revision 540
Max: Macro_ObjectsLights.mcr:
removed all references to an 'AmbientOnly' property that my PointLights don't
seem to have (?)Or that isn't
accessible in quite the same way (?)
Prevents MaxScript error(s) when doing 'unhide all' with a PointLight
selected.
Max: PointLight: now has a
'Reverse CullPlane' option so that the 'facing' cullplane has to be
backfacing for the light to draw.
Added this for the lights on the north side of Regent Street; couldn't see
another good way of culling them because their ranges are too big to be
blocked by the cullplane on the John Street side of those buildings.They now require that cullplane to be
backfacing (they start drawing as you approach Katie's Kabin / Voodoo).
Windows: Changing 'theme' in
Display Properties resets the 'sound scheme' to the default.Sound schemes aren't saved with the themes
either.So any time I change my
colours to read some text, I have to go and switch off all the default
sounds.Then I have to switch them off
again after changing my
colours back.
Had some CullPlane bugs, started
fixing them
TODO: FINISH THIS!,<-cpRegentNorthWest
was facing the wrong way, duh.
RELEASE v11
18-May-2007
0
Planned a field trip using
Expedia.This has been on my todo list
since 10.1.6 (the 500-hour mark)
17-May-2007
4
TODO: need to incorporate collision mesh generation into my
pipeline.At the moment,
CLandscape::GetGroundHeight is spinning through up to 16,160 triangles per call.The rejection is very efficient
(four-or-less int compares as a bounds check), but still...
CLandscape::GetGroundHeight now
just uses the first triangle it hits rather than finding the highest one
(what was I thinking there?!)
Trying to fix this collision
problem (Regent St. south pavement):
- not using D3DXComputeNormals; that rules out bad normal
extrapolation.
- All outer edges of the collision mesh are showing collision faults (only
some tris though)
- removing the triangle bounds check makes no difference
- when it fails, it's not hitting any triangle
<-Oops, I was still using some unsigned shorts despite having changed to
32-bit vert indices recently.Fixed
now.
added functions Vec3Abs, Max3
Still had a few tiny 'cracks' in
the collision.For example:
At these coords, 'alpha+beta' in IntersectVerticalRayGroundTriangle was
1.0431895554065704
X: -11432.433
Z: 3418.2939
Now changed the function to to compare alpha+beta with 1.0666f
(1+ALPHA_PLUS_BETA_MARGIN)
Haven't seen any problems since.If
the problem shows up again I should check the details of the triangle(s)
involved...
Max: Added a cutter object that
removes the undersides of the pavements.This was needed now that the ground collision returns the first hit rather than the highest
one.It also removed 3225 unecessary
ground triangles.
Version 11 is developing
nicely.Will release this soon.Exact date depends on when my rain deadline
is.
TODO: work out why the
north-side Regent St. buildings can cut out at certain viewpoints.CullPlane problem?<-the
cullplane wasn't a child of the building as it needed to be
16-May-2007
3
Added lights and kerbs on Regent
Street.Editing in Max was painfully
sluggish for some reason, but other than that the construction process is
working well.
CullPlanes are working well too.
TODO: work out why collision is
failing on that south pavement<-fixed (17th)
15-May-2007
2.666
#revision 531
Tracked down yesterday's Max
crash.It was the #ifndef, #define,
#endif wrapper in NoctambuleInclude.fx (#included from Patchwork.fx in
version 528 which caused the crash).
I have no idea why it would cause a problem; fxc never mentions it when I
compile.
It doesn't actually need to be there, so I've removed it now.
I think I'll get out of the habit of putting those wrappers into hlsl
headers now.
I'm very glad that header
was version-controlled!!!
Working on puddle improvements
still.Looking nice.
TODO: add loads more lights
now!!They don't have to be in the
right places yet, just need to get the right sorta number of them so the
lighting will be more realistic.<-done (16th)
This will open up Regent Street for v11, w00t!
14-May-2007
1.666
Parallax.fx, ground.fx: Removed
g_vTextureDims, not needed now.
Improving puddles.Making them less binary.
Max started refusing to open
the town (crashing).Luckily version
507 loads fine (after rolling-back the whole Noctambule folder)
TODO: find the latest working revision.<-all sorted now (15th)
Took down the brightness on my
monitor quite a bit to make it more like the one at work.
13-May-2007
7.08333
Max: Schematic view: 'include
only visible objects' seems to be broken at the moment; shows me just six
loose objects and nothing else (?)
I'm finding that POM heightmaps
benefit from being halved in res; makes the protrusion edges work more
easily.
Also makes the surfaces look wetter (gives the impression of water surface
tension I suppose).
REMEMBER: No E&C on patchwork.fx!!Still don't know why.
#revision 527
Reached a workable compromise
with ground POM: only patchwork height is now used.Dimple parallax was doing too much damage
to the kerb/kerbside texturing.Cleared away PatchworkDimpleHeightCombine.fx; last revision it appears
in is 527.
Ground.fx: now using a better
(hopefully correct) view vector for PSGetPOMTexCoords.This removes the squidging I was seeing
when close to the ground, phew!
Added GroundPOM.fx (just defines
GROUND_POM then includes Ground.fx)
Got the puddles refracting
(ground.fx), looks lovely.
- Not geometrically accurate (would be downright foolish to pay the cost
involved, considering that these are dirty puddles seen at night).What's important is that it offsets in the
right directions, takes into account perspective and water depth, and LOOKS
right.
- Borrows some data already calculated for the reflection distortion.
- Patchwork system, little gem that it is, eliminates the need for a screen
copy like you'd normally need when doing refraction.
- Works very nicely in conjunction with ground POM.Strengthens the illusion of depth, because
you can see the bases of the protrusions (deep) refracting farther than the
tips (shallow).
- Shadows are refracted by offsetting the ws position output.Doesn't match perfectly but I think it'll
pass.I think it's matching in
direction at least.
Checked that the game still runs
fine in 16bit colour.
Checked that it survives a change of colour depth (through Display
Properties)
DVD due back on Sunday the 20th;
they're open till 11 apparently.I
don't trust them though.
12-May-2007
9.25
Managed to remove two muls (and
some other junk) from the existing ground.fx PS code, by optimising the
clip-space UV stuff.
Didn't have to multiply the UV by W in the VS because that multiplication
seems to be included in the world-projection muls...(?)
(above) Spent the day getting
POM working on the ground.This
upgrades ground.fx to the status of Ludicrously Expensive.
Added PatchworkDimpleHeightCombine.fx which combines dimple height onto the
patchwork height buffer.This combined
height drives the POM.
The biggest problem with the
ground POM at the moment is that it pulls the kerb textures out of alignment
with the kerbs.Kerbsides need to have deep dimples so that
puddles form there.The puddles
themselves will help to hide the misalignment I suppose....
Maybe need to make mid-grey the
non-displaced height (currently it's white).
I'm not 100% won-over by this
ground POM yet.I expect it'll prove
its worth when more chunky ground textures go in.But that's if I can get it house-trained.See how it goes tomorrow.
I'd be very interested to see
what one of those newfangled model 4 cards makes of all this.
TODO: puddle refraction<-done (13th)
11-May-2007
4.4
Made good progress with the
ground POM (carefully restructuring the code so ground.fx & parallax.fx
share what they can, h_GroundPOM.fx).It isn't working yet though.Been getting some very weird compiler errors....
Remembered how well-behaved the
rain impacts are when you see them from a shallow (low) angle.Tasty.
It's just the downwards angle, looking straight against the normal, that
looks naff.
TODO: Needs to be more thin, broken, strandy.
10-May-2007
3.5
Got the rain splashes lighting
again.I'm not impressed with them at
the moment though; they need work.
TODO: get them presentable.
Got the brickwork looking
good.It was good to get some practice
at taming the POM, getting the wetness right, getting the right texture
sizes, etc.
Made new road textures.In the height map, using 'minimum' filter
and curves helps to give the impression of water gathering in the troughs of
the surface.
TODO: Use Jpegs wherever
possible.
TODO: fix Photoshop normal map
exporting action (currently goes wrong if the alpha channel is blank to begin
with).
9-May-2007
0
Oops, overtime.
8-May-2007
5.33333
DIMPLES: Remember - can't do
any bold contrasty height detail using dimples.The three height samples are ten
centimetres apart, so a spot of pure white on black would appear as three
separated blobs of extreme normal.
<-improved this a fair bit so bold stuff works better, and reduced
sample spacing.
Improved dimples; I think they
were sampling from the wrong ground position, because I'd approximated a
plane intersection vector.
Improved
CMeshFrame::PreparePatchworkDecalRecursive to account for the enclosed ground
now being removed.This fixes the
disappearing patchwork objects seen yesterday.Asserts if ground can't be found beneath
the object.
Reminder: for ground patchwork /
dimples:
local normal: positive world Y
local binormal: negative world Z
local tangent: positive world X
Made new brick textures.TODO: neaten, reduce
to 512... Comically, without realising it, I chose
brick-for-brick the same piece of wall that I used the first time round!
7-May-2007
4
Despite my new rule about not
doing any overtime, I went into work and spent seven hours today tracking
down a bug that I still don't understand.The deadline was non-flexible so I didn't have a choice; still, I didn't
mean to spend that much time on it :(
Realised that the smoothing on
my kerbs was causing crazy normals across large areas of the raised
ground.Fixed this by adding another
level to the bevel to 'absorb' the smoothing.As a result the kerbstones will all have a bit of an exaggerated
roundedness, but that doesn't really bother me.
Started making preparations for
the ground POM.Improved dimple
influence.
TODO: find out why the BC
pavement (decal) is cutting out.<- fixed (8th), see
CMeshFrame::PreparePatchworkDecalRecursive
Ground contained by buildings is
now Booleaned-out to reduce the number of polys and (very expensive)
pixels.This is driven by a single
line object, very neat :)
(above) improved kerbs, road
bow.New pavement texture is 512*512
and currently 100cm².Works well but I
wouldn't want to go any lower than that.
TODO: ground POM."It is time."<-done
(12th)
TODO: normal/height maps to use
new format (XYH) since Z is now extrapolated by ground.fx.XYH format also allows the possibility of
XYHA targas, if needed ( patchwork shader would use min(colour a, normal a)
for the normals ).
6-May-2007
13
Max: Can't put a WSM PatchDeform on a spline???
Max: Groups: Use Attach to add new members to the
group.
Max: PatchDeform gizmo on a group jumps to the centre of the
group's bounds.Since transform
type-in is unavailable for the gizmo, it has to be lined up with the patch by
eye.Shouldn't cause a problem though.
Max: Kerb lines are now in a
group.This allows each loop to be a
separate path for the benefit of PathDeform.<-now
using a Connect object containing all the pavements
KERBS: can only deform to coarse ground patch.This is because the final ground patch has
horizontal vert adjustments.
XForm modifier for scale (on a
patch) : BEWARE: This has been seen to cheekily bake itself in above the
modifier stack.Avoid using it where
possible.
Max: Lines: To ensure mesh kerbs
etc. match with path deforms: un-tick 'optimize' but
don't tick 'adaptive'.
No need to use Normalize Spline and it's best to avoid it for my purposes.
All kerbs are now smoothly
bevelled :)The highlights that this
creates along the edge of the kerb are quite valuable, and I can see the same
highlights in my reference photos.
Max: Learned about Groups, fell
out with groups, learned about Connects, Connects rock!
Made
some textures for the pavement outside Bedford Court.
(above) Smoothly bevelled kerbs,
new paving tile texture.
Spent all day sorting out Max
stuff, finding the best ways to proceed with building my town.Very time consuming work (and lots of hangs
and crashes) but I think it's really important to get it right early on.
Had an idea that maybe something
like an Edit Normals modifier on the ground patch could stop it deforming
ground verts horizontally.Instead it
broke the deformation altogether (and was quite chuggy).Edit Normals turns the patch into a mesh it
seems.Tried slapping a Turn To Patch
on top of that but it then told me it was an "Illegal patch" for
the deformation.
Tried all sorts of things to
solve this problem of PatchDeform shifting verts horizontally.Couldn't find a good solution.Could do some stuff using Conform, but it's
not very future-proof because anying Booleaning onto the ground has to do so
post-Conform.
All I need is a simplified version of PatchDeform that
only acts on the vertical axis.
Decided to continue using
PatchDeform as before.It's possible
(but pretty unlikely) that a simplified version of the modifier could be
found or written later on.Till then,
I'll just make do with the fact that changing the ground slopes will shift
kerbs and other Booleaned objects horizontally by a foot here and
there.
Might not be a big deal anyway - maybe I'm approaching this modelling lark
too much like a coder.
Just need to remember not to leave any crazy test
humps in the ground patch.
PatchDeform: This has a pretty
serious limitation in that it can't deform to a patch on which any splitting
of edges has been done (even if the splitting results in a perfectly clean
grid formation).
Workaround for the problem
mentioned above (splitting patch edges breaks PatchDeform):
To add more resolution to a patch without breaking PatchDeform:
- PatchDeform a new higher-res patch onto the existing one
- Collapse the new patch's stack; this turns it into an Editable
Patch
- Delete the old patch; deform the objects to the new patch.
I've tested this to make sure it works.
Another reason why Normalize
Spline is needed is so that long lines can deform accurately to the ground
patch.
Got the roads realistically
bowed now.'groundPatch' is a flat
patch apart from bulges down the road centres (and whatever else I want to
use it for).It's patch-deformed onto 'groundPatchCoarse'
which contains the slopes/hills.Got
rid of 'groundPlane'.The ground
chunks now Boolean straight out of 'groundPatch'.
I'm now deforming all kerbside
road lines by the coarse ground patch.
Made and applied textures for
the Bedford Court car park.Looks
great. Really needs a gloss channel though, hmm....
TODO: use a single meshified
reference of the kerb line for all patchwork Booleans, don't let them keep
private references as operands.<-done
TODO: work out how to
path-deform onto the second loop of the kerb line; if it's not possible then
the kerb Booleans need to stack up....<-now using a
Connect object containing all the pavements
4-May-2007
5
Managed to deform some paint
along the same line used to raise the kerbs, with W's help.
Found I had to apply a Normalize Spline modifier to
the line to get it to match the deformed paint.<-No,
just un-tick 'optimise' in the line's 'interpolation' panel.
Learned a bit more about using
references & instances, and how modifiers work with them.
The PatchDeform on the ground
mesh actually shifts vert positions slightly on the horizontal axes, which
was bringing the kerbs out of alignment with the ground patchwork.To fix this, I'm now putting the kerb lines
through the same PatchDeform, which works well but makes it slightly tricky
to position road lines up against the kerbs.
Kerb extrusion: can't un-tick
Cap Start (it would break the Boolean), so I have to make sure the 'start'
(base) is slightly under the ground.Otherwise the Boolean can get really puzzled, doing loads more work
than it needs to and making a horrible mess.Got several hangs & crashes resulting from that.
Kerbs: I'm sinking the extrusion
10cm into the ground, otherwise it doesn't reliably intersect the ground.
Tried using Vertex Weld to
stitch up the kerb joins; couldn't get it to work right.
TODO: Try lofting the kerbs and
using the Bevel deformation.<-now using Bevel
modifier, much easier
3-May-2007
5
Had a bit of a breakthrough:
realised I could use Booleans to split up the ground into flexible
(adjustable) chunks.
I had always been a bit worried by the question of how to split up the
ground mesh (if at all), but it looks like it's going to be a doddle.
Currently just using one chunk to save me having to make the changes to the
collision system just yet.
also...
I HAVE KERBS!!Flexible spline-based
kerbs (Boolean'd onto the ground plane pre-PatchDeform) are working
brilliantly.This is great.It's like Christmas this evening.
The collision system didn't much
like the Optimise modifier I tried putting on the ground chunk (polys too big
I suppose).
TODO: find out why this causes a problem.
Tried path-deforming some road
paint round the kerb lines (as a test to see how this'll work with kerb
textures).
Couldn't quite make it work because:
- the line needs its Extrude modifier to work in the Boolean with the
ground plane
- the line needs its Extrude removed/disabled to work in the Path Deform
for the paint
- can't seem to make one reference of the line have an Extrude and one not
(without losing the 'connection' between the two references).
I'll ask around and see if anyone has any ideas.
2-May-2007
4.83333
Experimented with possible
methods for road & kerb creation.Learned a lot more about using Booleans and other compound
objects.
Using a combination of stuff like Booleans, Patch
Deform and maybe little local FFDs to squish the kerbs, I should have no
problem at all keeping the ground mesh flexible, which is a big relief.
For the large-scale undulations of the ground, it'll probably be best to
use Patch Deform rather than an FFD (which wouldn't allow me to adjust the
resolution without resetting the deformation).
TODO: would be brilliant if the patchwork kerb
objects could drive the Boolean that raises the pavements to form the
kerbs.That way I won't have to worry
about keeping the patchwork kerbs aligned with the mesh kerbs (which is going
to be a nightmare otherwise).<-all sorted :)
Max: FFD: changing the number of
points resets the volume >:(
Max: Conform: To direct the
projection: "Activate the Top viewport. In the Parameters rollout >
Vertex Projection Direction group, choose Use Active Viewport, and click
Recalculate Projection"
This seems to be the easiest way of doing it.
Won't be using Conform for the ground though, because it flattens the
wrapped object (or certainly seems to whenever I've tried it).Also doesn't seem very user-friendly.
Started adding the proper
ground-shaping and road-cutting objects into the world.Going OK so far.Still not sure how I'm going to paint water
velocities onto the ground this way though (in a future-proof way).TODO: look for a
modifier that will allow this - surely there has to be one.
1-May-2007
5.25
#backup
Backed-up the repository and the
working folder onto DVDs at the end of last night, using DeepBurner.
Max: children-of-lines problem:
The lines aren't exporting, even though I'm asking the exporter for dummy
objects as well as geometry.As a
result, children of lines are jumping up to the root of the hierarchy where
they're no use to anyone.
What can I do?There are no other
options on the exporter; the lines aren't hidden in Max...
Just going to have to avoid making anything a child of
a line.
Max: double-clicking an object
selects it and all its children
Max: Lines can be children of
the objects deformed to them, but this connection has to be made before the Path Deform modifier is
applied, otherwise Max complains about cyclic dependencies.
Max: Schematic view: always move
nodes slightly before disconnecting them, otherwise they'll jump off to the
other side of the schema, even if 'always arrange' is off.
Road lines: Don't attempt any
overall transform on the splines themselves; this breaks the path
deformation.Move the points around
instead.
Got a Max crash when Windows was
expanding the virtual memory.Now
locked the virtual memory size so it hopefully won't do it again.
Max: Crash: Putting a
VertexPaint modifier onto a multi-selection of lines (by accident)
TODO: need to use the
non-lollipop lighting shape for reflected lighting.
Spent most of the evening
arranging the ground patchwork hierarchy.Needed doing - much nicer now.
Each road line is now the child of one of the decals deformed to it.
Found that it doesn't make any difference which object is the root object
of its area - doesn't have to be the biggest or most central object.That's because the frame's mesh sub-frame
has its own bounds like any other child.For example, 'generalLandscape' doesn't draw its little building (down
Augusta Place) unless that building is genuinely visible.
For that reason, I'm glad I have those weird empty container frames.
Thought about roads &
kerbs...
Played with cutting a road into
the ground using ShapeMerge.More
experimentation to be done here... (leamRoadTest.max)
30-Apr-2007
4.83333
MAX: any time a shader is
reloaded and doesn't compile, all of the material parameters are reset to
defaults.
Thanks Max.
That's why it's safest to check that the shader compiles before reloading
it in Max.
MAP SCALER:
- Unfortunately Map Scalers work relative to the object's edge not its pivot/centre.As a result, UV adjustments will be needed
on the object in most cases (use either the Map Scaler's offset controls or
an XForm UVW).NOTE: this adjustment depends only
on the shape of the object, not the position of the graphic (phew).
- Although Map Scaler isn't perfect in this respect, I think in most cases
it'll be preferable to using 'loose' UVs.
Max: Remembered about the FFD(box) modifier.This is ideal for distorting road
lines.
Set it to something like 2x32x2, and remember to pull pairs of control points around
because there's two layers of them vertically.
TODO: Max: Since path-deformed
objects are so difficult to select with the mouse, and for other reasons,
could do with a custom road-line modifier for the splines themselves.This could include:
- 'select all paint' button
- maybe some way of giving the paint random noisy vert colours & alpha?
TODO: Try using alpha thresholds
on the road paint, so it'll break up nicely according to its vertex alpha.
Grrrr, decals don't currently
show up in the game if they're children of lines.They do export, but do they get parsed
properly by D3DX?I'm hoping it's just
my code that's at fault.TODO: FIX
THIS!
Finally sorted out the Max UV
discrepancy (27th), properly this time.The h_generalPOM parameters are all doing their jobs properly now and
allow texture portions to be upgraded, repositioned and rescaled without any
edits being required on the objects themselves.
Patchworks: Had an idea that
certain patchwork objects could be given a tick to say "bake me and my
children onto a texture and replace us with a textured quad
accordingly".In this way, bits
of patchwork that are getting too busy can become texture-patchworks without
any faff.That will be a beautiful way
of working.
If I had all the time in the world, I'd probably want to make the game decide which objects to bake
based on how long they took to draw etc...but that's just crazy talk really.
(above) The new spline-based
road lines, plus dimple & puddle improvements, and the
minimal-screen-area ("lollipop") lighting.
Lollipop Lighting™.Because you're
worth it.
TODO: get the decals drawing
when they're arranged as children of lines.Otherwise the hierarchy will become a complete mess.<-can't (limitation of the X exporter).Managed to get the hierarchy laid out
neatly nevertheless.
29-Apr-2007
3.25
My biggest (only) loss of time
nowadays is the overtime I do for Work.I need to sort this out:
RULE: NO OVERTIME TILL THIS
RAIN DEMO HAS BEEN DONE.NO SNEAKING
INTO THE OFFICE AT WEEKENDS TO CATCH UP WITH WORK.IF IT SLIPS IT SLIPS.THIS PROJECT IS MORE IMPORTANT.
If I can get into the habit of
never doing any overtime ever, and keep that habit up till the end, the
benefit to the project will be massive.I need that time to achieve the level of quality that is planned.
Road lines: DON'T USE SCALE!!
28-Apr-2007
5
Doesn't look like Leamington is
on GetMapping's recently-flown 12.5cm res coverage :(If only they'd gone right a bit.
WS PathDeform: use X as the
deform axis for ground decals.
Max: Road lines: use a modifier
stack like this:
- Path Deform WSM (path
deform axis X.Set 'rotation' to 180°
if a flip is needed)
- [Edit Mesh], if totally necessary.Caution: topographical dependancy.
- VertexPaint (even
an unused VertexPaint modifier fixes uninitialised vert colours)
- [Turn to Mesh]
- FFD(box).Set it to something line 2x32x2.
- Noise (Y axis,
for slight wiggle)
- UVW Xform (mostly for t-page offset, might
use scale in some cases though)
- Map Scaler OSM
(scale 1m, no offset, wrap=no)
- Plane (long
'width' and lots of 'width' segments)
TODO: Scaffolding (if there is
any): try using a LATTICE modifier, which basically makes scaffolding for
you.
Max: Found the Visual MaxScript
editor thing in the Utilities panel ('More...')
Should try using that for interfaces next time.
Max: Lines: Remove nasty sharp
joins using the 'fillet' control.
Patchwork decals can now use
vertex RGBA (patchwork.fx)
(below) Added some smashing road
lines, winding all the way down John Street.The process above works brilliantly.Adjusting road lines
is now literally as easy as editing a spline :)
TODO: paint-up the car park a bit.
Max: Sticking a VertexPaint
modifier onto an object (without even using it) seems to magically fix
uninitialised vert colours - yay!I
did it on a hunch and it worked.
Max: Don't un-tick split mesh on a displace approx modifier.Max will run out of memory and crash.
Photoshop: Learned how to use
layer masks.Handy!
27-Apr-2007
5.16667
Added a doc for my texture
production rules, otherwise I'll forget them:
TODO: puddle edges wobbling
slightly as they ripple in the wind?
Fixed a bug in
CLight::TestVisibility which was allowing nearly all lights behind the camera
to draw.Nothing sinister, just me
being silly.Got a tonne of frame time
back, phew!
Found & fixed another CRAZY
mapping discrepancy (this one is Max's fault).I just hope it happens consistently rather
than being a bit random.
See h_GeneralPOM.fx \ VSProcessUV, #ifdef _3DSMAX
Max: Found out about the WS PathDeform modifier which is
ideal for road lines...TODO: put a
bunch of road lines in.
Made a combined road-paint
texture as a (future-proof) placeholder till the texture patchworks go in.
TODO: work out where 'posters'
fit into the building patchwork / texture patchwork system.
26-Apr-2007
4
Fixed yesterday's UV mapping
discrepancy:
CTexture::Create now uses D3DX_DEFAULT_NONPOW2 instead
of D3DX_DEFAULT as the Width and Height parameters for
D3DXCreateTextureFromFileEx.
The bug was made more confusing by the fact that I thought my texture was
1024*256 when it was actually 1024*156.
Started planning my workflow for
texture production.The conventional
process of packing graphics together onto a composite texture is sensible but
seems faffy and makes it difficult to upgrade each graphic afterwards.So here's the plan:
texture patchworks (procedural
textures): all the 't-paging' is done using geometry in
Max.Every texture file will just have
a single graphic on it, and there's no faffing about with resizing.I can then change individual graphics
fire-and-forget style.Importantly,
the texture-patchwork approach makes it easy and efficient to create variants
of textures (using decals same as on the ground).
Normal/height-map generation for texture patchworks will be done at game
startup using a material on the texture-patchwork component graphics that
specifies the WS width & height of the graphic.
Here's the difference between
patchworks & texture-patchworks:
Patchworks
are rendered view-dependently on the fly and are only suitable for plane-ish
surfaces.They'll mostly be used on
buildings (as well as the ground) and will probably team-up well with
CullPlanes.When they're small on
screen, they can freeze 'as-last-projected', be held on a small ORT and
continue to be mapped correctly onto the surface.This might have to sync with a switch of
material technique for the receiving mesh.Imagine the efficiency of freezing them like this!!A whole side of a street sitting on a
single perspective-inclusive little texture.
Unlike the ground patchwork, building patchworks will be applied using
explicit uv coords (Unwrap UVW).This
is for continuity around doorways, alcoves etc.Do need to make sure that the overall
mapping lines up with a planar map though, otherwise we'll see bad
screen-edge artefacts.
Texture patchworks are rendered orthographically (top view), the first the time
the game is run (during install would be nice) and then saved as
textures.They'll be used on patchwork
decals and general objects.Texture
patchworks can be multi-textured onto an object to provide multiple layers of
detail ranging from say, a crisp h-tiling brick formation to low-res muck and
brick-colour variation.
TODO?: Can the kerbs maybe use a
sort of mirrorred-mapping trick where the sampling pos moves back along the
outwards kerb normal?They can be
flagged as kerbs using channel mapping.Still have the awkwardness of lining-up the kerb texture with the edge
on the mesh though.
Er, didn't get much rain done tonight :/TODO: make that
combined road-paint texture as a (future-proof) placeholder till the texture
patchworks go in.(<-DONE)Then do some rain!
25-Apr-2007
3.5
Patchwork now uses MRT again
like it originally did.
patchworkColourRT and patchworkNormalsRT are now replaced
by groundPatchworkRT
I've got a weird mapping
discrepancy between Max and the game, on these road lines.The material properties are coming through
fine to the shader code; it's as if maybe the exported UVs are wrong.Happens even without any mapping
modifiers.TODO:
FIX THIS<-FIXED (26th)
Patchwork height now comes
through to ground.fx and contributes to puddle depth.
24-Apr-2007
4.16667
MAX: Found out about the WS Map
Scaler modifier.Thought this might
fix my old mapping problem, but it doesn't.
MAX: Create > AEC Objects
> Wall.Handy!
Need a way for some patchwork
decals to be colour only or normals/height only.TODO: maybe separate colour/normal alpha
multipliers in the material?Maybe
colour-only and normals-only versions of the material.
ImageAlign Pro no longer works
(wants me to register).That's
annoying.TODO: find out how to
achieve the same stuff in Photoshop.
<- LENS CORRECTION!RAR!That's about the best thing ever.ImageAlign who?
(below) Made a proper
double-yellow lines texture.Quite an
encouraging result - you can see that it's going to look great when I start
putting all the proper road markings in.
Had a think about wall/object
patchworks, having kinda realised that using conventional texturing methods
on the buildings would involve an awful lot of learning and practice (and
probably more time than I've got).Very
keen on patchworks, will decide exactly what I'm doing with them tomorrow.
21-Apr-2007
4.83333
Noctambule.sln had no projects
in it when I opened it this evening (thereby differing from last night's
check-in) - how on earth did that happen?Now reverted and ok.
Reloading has stopped working
on Patchwork.fx.Surely this has got
to be something to do with the textures having moved????
NOTE: POM material properties
such as texture dimensions and texture aspect ratio, they really are for the
TEXTURE not the graphic.A material
using a 100cm square graphic occupying one quarter of a texture should specify
that the texture WS dimensions are 200*200cm.For a 70cm square graphic occupying a quarter of the same texture, the
material would specify that the texture was 140*140cm.
Really need to compile these rules into a doc (notes are gathered in the
file below)
Decided against the idea of
patchwork heights influencing puddle flow (ie. bumps casting proper little
wakes into the stream).The number of
texture samples I'd need to get it looking good would be [very] excessive given
the relatively low visibility of the effect.The velocity map results weren't far off, so I'll go back to that.
Had a long think about the
scale-type properties on these POM materials (planning how the ground will
work with POM).
All that's needed in order to sync the normals with the depth is:
- a ratio between the texture width (NM height scale) and the desired
depth
- an aspect ratio between the texture width (NM height scale) and the
texture height.
In some cases I'll know the actual WS dimensions of the texture and fill-in
the material properties accordingly.The rest of the time, I'll just leave the WS dimensions at the default
of 100*100 (if the texture's square, else...) - the depth property is then
just a percentage.No harm in working
that way at all (and no awkward changes required for existing patchwork
objects).
Irregularly-stretched UV mapping is the only thing to avoid, but I don't
have much use for that anyway.*Regular* UV scale works a-ok; the depth scales with it.
TODO: get to the bottom of that
weird UVW Map scale problem I always get in Max (get road line objects
properly mapped)
20-Apr-2007
3.75
Added Photoshop action Export Normal+Height Map to export
a normal+height tga from a layered height-map, via the optimal 16-bit route.
All height PSDs will now live
next to their TGA output.Started
setting up a proper texture folder hierarchy, to facilitate batch processing
later on (and just to help keep track of the textures).
* From now on all GROUND normal
maps will be generated using the same rules for normal-map scale as the wall
POM maps (SCALE IS THE TEXTURE X RESOLUTION).
* I'll add some multipliers to the patchwork material, probably the exact
same ones as parallax.fx.These will
have to be adjusted whenever the textures are combined-up (t-paged), that
can't be avoided.Make this adjustment
at the same time as applying the UVW XForm modifier to each material involved
in the change - that way there's no guesswork involved.
* Ground will use POM (after the rain deadline), and this will be
beautiful.The POM caveat about
joins/decals disappears totally when using the patchwork system (YAY
PATCHWORK SYSTEM!!) and this is one of the reasons why I keep thinking about
wall patchworks too...
19-Apr-2007
4
Max: Grr, texture can't be shown
in Edit UVWs interface for DX9 shader materials.
Max: UVW
XFORM modifier is going to be my friend.With this, I can initially texture objects
using loads of individual prototype textures, then *later* t-page them up any
which way I feel like.I'm guessing
this might be common practice, but it's all new and clever to me!
(Use 'select by material' in the material editor, add a shared UVW Xform
modifier onto this multi-selection)
TODO: Make Patchwork.fx use
MRTs again - will be a load faster (made it non-MRT as a build bodge for
compatibility on an old nVidia 5xxx I think)<-DONE
25th
Dimples now support vertex
colour & alpha.
Did a load of dimple
experimentation in Max - checking I haven't overlooked anything major.All looks good.
TODO??: would like a way of
controlling the degree to which the peaks of each dimple material smooth the
patchwork heights&normals.Could have this as an RGB channel on the
dimples RT (ground.fx can read it straight from there), but this means adding
another sampler just for that purpose.Alternative would be to steal an RGB channel from the dimple normals
RT and reconstruct it in ground.fx.Neither sounds good.
NOTE: This is a pretty minor wishlist thing so don't bother if it's going
to have a significant cost.
Max: Found out about the VertexPaint modifier, which is a
great way of setting or painting vertex colours (or channels) across multiple
objects.
TODO: Finish getting patchwork
heights written so they can influence puddles and puddle flow (patchwork.fx).Er, I suppose this
will involve stealing the patchwork normals Z channel and paying a sqrt to
reconstruct it in ground.fx.Needs to
be done one way or another though.<-DONE 25th
18-Apr-2007
0
en plus de ca
Reserved bebo pages in case I
want them at some point in the future:
lecauchemar.bebo.com
programmerart.bebo.com
17-Apr-2007
4
Added
CLandscape::groundEffectPlaneMatrix.TODO: remove point & normal.
Got dimples working properly.They're great, and
already give that car park the same ruggedness it has in real life.Now uses three texture samples per pixel
instead of 8.These three height
samples are offset from each other by world-space vectors now, so the normals
are consistent regardless of viewing angle & viewing distance.
TODO: light marker sprites are
only drawing to one stereo eye; fix this.
TODO: test stereoscopy with
convergence=1.Doesn't look the way I
remember it...
TODO: back to flowing puddles,
using patchwork heights...
16-Apr-2007
0
work
15-Apr-2007
0
work
14-Apr-2007
2.58333
Changed meshes back from using
32-bit vert indices, because it was causing some holes in the ground
collision (absolutely no idea why).
I expect I'll have to fix it at some point, but not now.Search for D3DXMESH_32BIT - I've left all
these in but remmed them out.
une question qui brûle les
levres
Max: After rearranging the
dimple hierarchy (so the tile layer is the root), the dimple objects were all
distorted.Fixed them by
multi-selecting all of them and just clicking 'reset transform', nothing
else.
Patchwork and dimples
all drawing again now.
CMesh: Added member decalOrderingY (see comment)
because centre positions are shifted up to the ground collision and therefore
can't be used for ordering.CMeshFrame::QSortCallback_VerticalDepth now sorts using this.
Multi-monitor: Would be very
cool to support multiple monitors properly.From glancing at a DX FAQ it looks like each monitor gets a separate
adapter object and you create a device for each of those adapters.
Changed meshes to use 32-bit
vert indices (needed when testing a subdivide modifier on the ground mesh).<-changed them back; see 14th
Spent most of the evening trying
to tune the flowing puddles after changing them to follow the ground
velocities.Didn't manage to get it
looking as sweet as it should be...Last
night's results were pretty flukey I think, and I don't like effects that
work by fluke, for this very reason that they're difficult to tame - I could
spend forever tweaking these numbers and it might never look quite right.
So I'm going to step back a bit
and try to work out how to base the ripples on the patchwork heights because
that'll be pure gold if I can get it working.This should make the velocity map redundant, and produce the same
magic but better.
The deeper the puddles are, the less they'll be affected by patchwork
heights and the more they'll be affected by wind.
Refraction of the ground beneath will be the icing on the cake, and it's
made really simple (and pratically free) by the patchwork system (YAY
PATCHWORK SYSTEM!!)
X m'a fait cadeau de Y
12-Apr-2007
3
MAX: Added a 'Materials' layer
to contain dummy objects that keep hold of materials (Max material libraries
are not compatible with DX9 shader materials).
Started making the puddles flow
using a velocity map.This looked
gorgeous straight away, and I'm quite excited about it.I'm probably going to apply this technique
over the whole scene...
Started painting per-vertex
water velocities into the ground mesh using vert colours.These will combine with the tiling velocity
map.
TODO: get flow & impact
ripples combined again (ground.fx).Combine with per-vertex ground velocity.<-done
11-Apr-2007
4.16667
en garde à vue - in custody
au garde-à-vous - standing to attention
en première ligne - on the front line
à vol d'oiseau - as the crow flies
une vue à vol d'oiseau - bird's-eye view
un terrain de foot - football pitch
hors terrain
- off the ground
terrain de jeu - playing field / playground
sur le terrain - in the field
lancer une bouteille à la mer
SVN: removed .suo files because
they were making a nuisance of themselves following a rewind.
SVN: Close the devlog before
doing a rewind!!!
Made normal maps that will
scroll over/into each other to provide a foundation of wind-blown ripples in
the puddles (and possibly also distort the shape of the impact ripples).The *impact* ripples are now relatively
small and quick.Their job is just to
reinforce the impact splashes.Being
small and quick means they're cheap and they can now be used on the
non-puddle ground (everywhere) too, which will add a lot.
Got a new favourite puddle
reference video which concurs strongly with my choices regarding rain impact
and rippling puddle effects :)Gives
me a nice feeling of security that does.
10-Apr-2007
3.83333
Normal service resumes.Got as little as a month and a half left to
do these weather effects.
Removed all the fxo's from
source control, they were getting cumbersome.
TODO:
CMeshFrame::PreparePatchworkDecalRecursive is currently causing a problem
(changing the decals' Y order = draw order).FIX THIS<-done (14th)
TODO:
CMeshFrame::QSortCallback_VerticalDepth needs to sort by frames' OWN
centre/bottom pos (not centre/bottom pos including children).<-done (14th)
TODO: change dimples to
generate a world-space normals image, not ground-effect-plane-space.NOTE: dimple normals currently only affect
reflections, not lighting!Reduce the
number of samples used to create the normal map.Dimple normals only need to be coarse.<-DONE
Think I've worked out how to
make the ripples look good.TODO: get
this working.
9-Apr-2007
0
Work-work
8-Apr-2007
0
Work-work
7-Apr-2007
0
Work-work, although I didn't get
much done unfortunately.
à vie - for
life
le rendu -
rendering
Roughed-out my CV in shaky
French.Pretty happy with the form
I've chosen (basically listing what I did on each game that I've worked
on).I'm also trying to imitate the
ANPE examples to prevent it looking clumsy/unconventional to the person
reading it overseas.
6-Apr-2007
0
Work-work
rédiger -
write-up, compose
agir - act
(behave) / take effect.Lots of
compound forms to learn.
auprès de -
near / with / at / in collaboration with
un atout -
asset
convenir à
- suit, fit, agree with
convenir que
- agree that
au format PNG
un brouillon
- draft (rough copy)
soigner autant X que Y - give X as much care as Y
un réalisation - accomplishment, achievement
Read through anpe.fr's guide to
writing a CV, and grabbed their examples (C:\Noctambule\CV).
Will start on that quite soon.I
think it wants to be a web page, for easy viewing on the site.
Discovered the forums on
wordreference.com.Looks like there's
some really useful translation-type discussions to be found there.
5-Apr-2007
0
Catching up on work work.This is necessary to get morale back up to
a sensible level.Normal service will
resume shortly.
Found some great general-rain
reference in Hollow Man 2 (1:19:46 -> 1:23:11)
TODO: Try adding a couple of layers of fine scrolling speckles/noise into
the rain impacts image?
Added a 'PowerDVD Grab'
Photoshop action.Crops the image and
fixes the aspect ratio.
4-Apr-2007
0
ill
3-Apr-2007
0
ill
2-Apr-2007
0
Holiday
1-Apr-2007
0
Holiday
31-Mar-2007
0
Holiday
30-Mar-2007
0
Mr. O's leaving doo, he's off to
Paris.The way he described the
application/interview process made it sound way too easy.
Told me about other vacancies there too.Very tempting, but I don't feel ready yet :/I know I'm being cowardly.
Mr. B offered to help out by putting in a good word for me at
Montpellier...I need to visit these
places - the sooner the better.
29-Mar-2007
3
à la toute fin - at the very end, right at the end
ça a pas mal changé - it's changed quite a bit
Had another look at Empire of
the Wolves which has some excellent rain reference.
DOCS: Added reference.txt, for
keeping note of DVD reference footage (time points).
Tweaked rain impact splashes to
look a lot more dense.They were too
laid back, as were the ripples - these need to be very busy-looking (see EOTW
0:44:00).
The rain needs to feel unpleasantly severe, rather
than decorative.It should feel like
it's attacking the player, so that finding shelter feels rewarding.VERY important to appropriately change the
rain & wind sounds when in shelter.Done properly, this should affect the player at the instinctive level.
TODO: rain-in-the-eyes effect when you look up / sprint / face into the
wind?
TODO: wind affects player balance and movement speed?
28-Mar-2007
3.33
ça a l'air de dire que - that seems to suggest that
je m'appercevais que - I found that / noticed that
STRING TABLE / OPTIONS:
- won't go down the ultra-automated route.Options will continue to be manually parsed and manually added to the
menus.
- string table will not contain any menu-item data such as types, ranges
and visibility conditions.It's just a
collection of strings.
- option string names will contain the corresponding option member name,
for my benefit.Eg.
"GFX_shadows_destRT".
- option string names will not be case-sensitive
- options don't have to be grouped in the same way they are in the profile
files.
- enumeration strings will be named exactly like the enumeration entries,
but with the prefix "ENUM_", eg.
"ENUM_ESTEREOSCOPYMODE_NONE".They will be gathered and used automatically.The string names will replace the likes of
COptions::s_stereoscopyModeNames.
- enumeration values will appear in the order used in the string table, not
the enumeration's order.
- menu strings will be kept short, and information tips will give full
explanations.
Bit more work on the string
table file.
Noticed that fallback shadows
make the non-specular rain splashes disappear.TODO: fix this!!<-fixed,
just a colour-write conflict
27-Mar-2007
4.25
Considered using a triangle list
for the ripple sprites, decided to use quads linked by degenerates
instead.Simpler for me, probably
(academically) less work for the gpu.
For future reference, if the sprite graphic fits into a circle:
The right-angled sides of the triangle are the width of the quad *
1.7071067811865475244008443621048.
The triangle area is 1.4571067811865475244008443621048.
So about 1.5 times the number of pixels, half the number of verts.
au mieux de sa forme
D3DFVF_TEX0 MEANS **NO TEXTURE
COORDINATES**.IT DOES NOT RELATE IN
ANY WAY TO TEXCOORD0.
D3DFVF_TEX1 MEANS **ONE SET OF TEXTURE COORDINATES**.
Remember: One quad (joining
either side using degenerates) is 6 triangles in a triangle strip.Not 8.
AND YOU SHOULD DRAW FROM VERT INDEX [1] AND SUBTRACT TWO FROM THE TRIANGLE
COUNT!
Finally got the puddles
rippling.Looks sweet.Took all evening to get those sprites to
behave though.Mostly me being clumsy
with vertex formats...
26-Mar-2007
0
Lazy film nite, ice-cream and
all.
25-Mar-2007
6.66667
enlever -
remove (or kidnap)
un film à la Mad Max
merci d'avance
TODO: for lights & fallback
shadows:
- anisotropic specular lighting
- cross-fade between projector & fallback shadows
- fade aft lights in & out with visibility
- place more lights
Decided to leave the above until
I've worked on puddles & dimples.
Ground.fx: got rid of the stripy
artefact (changed normals sampler to 'point' filtering)
Made good progress with the rain
ripples.Added a very rough dimple
base layer, TODO: improve
Can't seem to pipe colours or
any other data into a vertex shader when drawing point sprites.I get the impression only position and UV
come through (generated by DX).
TODO: Change rain impact dots & ripples to use
manually-plotted sprites now :(Consider triangle list.
NOTE: Colours DO come through when using fixed-function pipeline.
24-Mar-2007
4
Fixed the fallback shadow
problem case (22nd) using depth feathering.(The ceiling in the BT pillar area was putting a hard bottom edge on
those highlights.)
Fixed viewport clipping for
lighting verts.
Specular tail is now included in area count (for
projector priority)<-nah, less jumpy without
this
le format
exigeant -
demanding, exacting, strict, fussy
monter sur scène / en scène,
quitter la scène - go on stage, leave the stage
épars(e) -
sparse
un sous-produit - by-product
Crikey, the 'useLightRange'
problem (ie. not working) in DeferredLighting.fx was just that I was using a
struct {float,float} instead of a float2.From this and previous experiences, it seems a good rule of thumb is:
Don't use even vaguely fancy syntax in shader code -
it probably won't compile properly.
Got the light mask textures
working nicely.These increase the
urgency of improving DeferredLighting.fx to give anisotropic specular
highlights.
Improved light visibility
testing to take into account the specular tails.
(above left) The signpost
reflection/shadow on the wall is a nice by-product of the fallback
shadows.It's unreliable though, and
is glitchy in many cases.Too much
glare here, obviously.
(Above right) Fallback shadows again, and the new feathering fixes clipped
edges previously caused by that low ceiling.
23-Mar-2007
2.666
Memory leak detection doesn't
seem to be working at the moment (tested with CProjector::s_instances -
changed from g_projectors).
Related to the above, I seem to
be using this convention for the 'p' prefix:
CItem someItems[10]
CItem* someItems
CItem* pSomeItem
CItem* pSomeItems[10]
CItem** pSomeItems
Added option
'shadows\numProjectors'.Made sure
this can be safely set to 0.
22-Mar-2007
3
Late start because I had to faff
with some stuff at work.
From now on I'll be getting in the habit of starting no later than 7 on
weekdays - guarantees four hours a day.
From now on I'm going to make a
habit of keeping the radio on while I work, to help with the French.I reckon it'll distract me less & less
the more I get used to it.I'm also
going to jot down words here in the devlog to help me remember them.Notated in whichever syntax I find most
convenient/helpful case-by-case.
Started fixing a problem case
with the fallback shadows, TODO get this working (in
GenerateFallbackShadowImage)
<-(23rd) nah, the fix was as bad as the glitch.Did a bit of tightening-up to the way they
work though.
<-(24th) fixed nicely using depth feathering
21-Mar-2007
4.333
Remember: Two disconnected quads
(joined by a degenerate) are 8
triangles in a triangle strip.Not 6.
Last night's work turned out to
be no good because I had overlooked the fact that one of the vectors, relying
on the result, needed to be normalised.The outcome of that is here:
I can't solve this specular tail
plotting problem.It has beaten
me.I just don't have the experience
in maths to know how to approach this one.There are people at work who could fix it for me, but that doesn't get
to the root of the problem and I'm sure they've got better things to be doing
anyway.
I've spent far too long on it already, so I'm going to bodge it, move on,
and start studying maths in my free time as soon as I get free time (which
will be at the very least several years from now).
In the mean time, I feel like a complete failure.
20-Mar-2007
1.75
I think I've sussed the equation
for the specular tail corner positions now, but it took an embarrassing
amount of time to get there.I wish I
had time to learn more maths (this is definitely on my todo list somewhere).
Thanks to MB for advice.
19-Mar-2007
2
SVN: head revision is currently
0.99/1.01 GB; repository is currently 975/978 MB.
Need a bigger memory stick and/or some way of burning DVDs.
Tried to calculate the corner
positions for the specular tails.But
it's one of those equations that doesn't wan't to be solved for the variable
I'm interested in (the reflection position).
Need to talk to someone who knows more about maths to find out how to deal
with this kind of problem.By all
means I could step through a range of values and use the one that fits most
closely, but surely there's a better way.
18-Mar-2007
9.25
Remembered about
SEnvironmentRenderData::pOriginalRenderTarget
fallback shadow image:
Firstly form a global ground-fitted reflection
position image (small RT):<-done
- draw ground with un-distorted reflection
position image glued into place on it<-done
- where ground is z-blocked, mask value is
written (fallback shadow can't appear here).<-nah,
works better without masking (wall tops etc. need these shadows too).
Form this image once the environment has drawn
but before it has been lit.<-done
Z-testing done manually using position buffer.<-works better without masking
renamed 'pvp' ->
'cMainViewpoint' to prevent confusion.It's NOT always the player VP!It's the principal viewpoint that is seen in the final frame (ie.
never a reflection/projector viewpoint).
Added CCamera::GetDirection and
CViewInfo::pMainViewDir (complements CViewInfo::pMainViewpoint).
Added CViewInfo::pMainCamera and
SetMainCamera, which records the variables above.
Note this is only called in one place: CModeEnvironment::RenderEye.
Lost about 6 hours by going out
last night (up too late and tired this morning).It's nice to socialise etc, but I need to
remember that it does take a
hefty bite out of my schedule.
Got fallback shadows behaving
nicely.They're sweet.
They're only used to block specular.
Need to fade as discreetly as possible from projector shadows to fallbacks
as you walk around.Technically the
cross-fade will be easy, just hope it's not too obvious.
I think I've got too used to
looking at these fallback shadows - the perspective on the real ones looks
weird now.
TODO: Disabling landscape
backface culling improves the fallback shadows in some cases (where a
building should be occluding a highlight but the highlight is higher than the
ground...)
So maybe include a landscape backface culling option (currently always
backface-culled).
OR, build downwards-facing 'floor' polys into the buildings??That would be cheaper.
NOTE: Parallax.fx currently doesn't appear in backfaces, ie. it hides them.
TODO: Use ProjectorShadow on the
lighting geometry, rather than covering the whole RT.It's not a cheap shader<-done
Projector: TODO: use a minimised
(narrowed) version of the view frustum which points at, and fits around, the
light's range sphere?
TODO: Finish specular tails
17-Mar-2007
8.41667
Max: Noticed you can drag a
material onto an individual selected poly?Experimentation needed there.
NOTE: Rotation is not yet
supported on collision rectangles - don't use it!
Did a bit of
modelling/placement: walls, cars, collision
ATI QueryLightFlare sample
(debug) drops from 16/1700 fps to 13/1400 fps when you reduce #queries from
>1 to 1.
Would stand to reason that it would be far more severe in a detailed scene,
but I haven't tested this.
This is just to reiterate what the doc says, that hanging around waiting
for recent (this frame's) queries to finish is a bad idea.
Realised that the reflection
position buffer can be used to form a pretty respectable fallback shadow
image for lights without projectors.Began setting this up...
SFallbackShadowVert: Couldn't
get D3DFVF_NORMAL (VEC3) to work for the light position, no idea why.
D3DFVF_TEXCOORDSIZE3(0) looked more like it was working but it
wasn't........??????
16-Mar-2007
2.75
The nVidia Quadro card in the
Vista machine at work exhibits the same zbuffer artefacts that M gets on
7900.Will remote-debug this some
time.Also suffers blackouts like the
rest of the nv cards.Can fix these
two bugs together.
Added
CLight::GetProjectorPriority, which is really doing the business now.Mostly thanks to 'mainDrawClipSpaceArea' which is a great indication
of how much a light needs a projector.
Number of projectors currently set to just 3, yet it's enough to cover the scene
nicely 90% of the time :)
TODO: Fix the 'useMaskTexture'
thing in DeferredLighting.fx
15-Mar-2007
5
VEC3: Added wrappers Vec3Add and
Vec3Sub.Because 'D3DXVec3' is quite
an obstacle course.Also good to have
consistency.
Maths: Added
GetClipSpacePosition and GetClipSpaceSphereDimensions.Found that these aren't appropriate for
working out lights' screen bounds, but they work as intended and they're
handy functions to have.
Also moved-in GetDistanceAlongLineOfSight from Util.
Maths: Added
GetClipSpaceSphereCoverQuad.The calcs
were slightly wrong (couldn't figure out how), so I simplified them to
something that is perfectly suitable for the lighting bounds (and far less
wonky).
The version that I thought *should* have worked is in stuff.cpp.
Upped the number of projectors,
and was reminded how much rendering work each one does :|Second pass of the shadows should
definitely finish with an optimisation spree.
TODO: Arrange projectors so they only generate their depth images for the
first of the two eyes.They do each
have their own face RT's in readiness for this.NOTE: common depth images but unique shadow
images.
Lighting 'head' quads now
plotting nicely.Remember that these
fit the light's bounding sphere
not their range.
TODO: apply mask texture.<-done.Mask texture
works an absolute treat.
added lightMask.png
14-Mar-2007
1.83333
CProjector::GenerateDepthImages
now uses a private z-buffer.This
allows light marker sprites to draw after the lighting (main z-buffer would
previously have been mangled by the projectors by that point).
Requesting D3DFMT_D32 returned D3DERR_INVALIDCALL; now using D3DFMT_D16
instead.
debug rendertext prevents
alt-enter TODO fix<-done
Got light marker sprites drawing
properly at last, phew.
13-Mar-2007
6
#revision 423
**** ALWAYS USE FOUR-ELEMENT
VERTEX POSITIONS ****Vertex declarations seem to expect this, even if you tell them
FLOAT3.
Vert types with vertex
declarations can store their members in any order, as I'd expect.
Setting a vertex declaration
causes it to take over from the current FVF.No need to use DEVICE->SetFVF(0).
DON'T USE ANY POINT SPRITES IN
THE GAME SCENE.They're not aspect-ratio-friendly and they're basically unsuitable for
world-space stuff.
If you do want them to scale automatically for perspective, you have to set
up D3DRS_POINTSCALE_A, D3DRS_POINTSCALE_B and D3DRS_POINTSCALE_C, which are screen-space sizes.Also set D3DRS_POINTSCALEENABLE.
D3D: No quad-list prim type??
v10.2 zbuffer mess: Looks
exactly like what you get if you write to the zbuffer but not the position
buffer.
TODO: Check that the position RT will definitely be set when the zombies
draw.
Spent the whole evening getting
these placeholder light marker sprites working.Took a long time because it's the first
sprite stuff I've done in DX.Learned
plenty from it though.
Working now, just need to find a good point to perform the draw.
12-Mar-2007
4.66667
Lighting geometry can't benefit
from z-testing (objects in front of the bounds always need to be able to
reflect specular).
NOTE: All the functions in
Debug.cpp take vector references rather than pointers.This is because it's handier for when you want to pass in say this->position+this->normal*100,
not having to wrap stuff up in brackets and ampersands.
Everywhere else in the code, I'm sticking to vector pointers which matches
with DXUT.
Got debug spheres working
again.Now used for light debug.Now take a colour param so I can tell
different lights apart.
Put a load more lights in.This is great because it has the effect of
opening up new scenery to explore (albeit shoebox scenery).v11 should look a lot more impressive than
v10.
Really impressed by the cleanness of the shadows cast by small lights close
to the viewpoint (eg. Bedford Court west wall).
Improved sequencing of projector
work - they were a frame behind.Also
improved allocation of projectors to lights TODO:
improve this.<-done
TODO: Get that lighting geometry
plotted!
Realised that the specular 'tail' isn't necessarily connected to the head
part, but I do need to make sure they don't overlap.
Noticed that a big part of
streetlight bounds can be covering empty sky.TODO: use stencil testing to reject these pixels before they get to
the PS (I assume that's the order?)
TODO: Streetlights don't really
cast upwards.Surely I can use that to
optimise the shape of their lighting geometry.
11-Mar-2007
5.91667
Got some flickering lighting by
accident...Realised that
electric-blue sparking lightsources will look great in dark areas as a
complement to the peachy/orange streetlights.Maybe the enemies should spawn The Terminator Way or have some kind of
electrical effects associated with them?
I've got blue energy weapons in my head too.Blue light is very sci-fi and
other-worldly.
Also, streetlights cutting out
as the enemies approach, that could be really scary.The more the enemies and weapons can mess
with the environment the better.
Can't breakpoint 'goto' lines,
even in a debug build.
NWF: Hidden lights no longer
export.Checked this works ok with
referenced light nodes.
Added CCullPlaneUser which
meshes and lights now inherit from.Characters will do as well.
Lights now have sensible culling.
10-Mar-2007
8.5
This weekend I'm doing/starting
the second pass on the lighting (not fixing the shadows, that's a separate
task).
I thought it was important to
get the lighting more decent before carrying on with the rain effects.Otherwise I'd be constantly asking myself
"yeah but how will this look when there's more lights in the scene?".
After lighting, I'll do puddles and ripples.Then the falling rain.
Had a little think about how
lighting and reflections want to play together.
Decided that neither one risks becoming redundant, but that each one should
be able to pull its own weight if the other one was switched off.Providing options for the number of lights
and reflections will be grand.
The indications so far are that the two effects are good at covering each
other's faults, and they certainly don't conflict in any way.A light object with a corresponding actual
light can be hidden from reflections (or reflected un-luminous) if that looks
better (eg. because reflections can't cast shadows).
TODO: try running the blurred
shadow image through a squaring curve:
(below) Planned the geometry for
the lighting.Lollipop shapes made
from a square 'head' (encompassing WS bounds*0.XXf, diffuse+specular) and a
tall 'tail' (specular).The shape will
be feathered by a single mask texture matching the lollipop shape.Specular tail width and height are governed
by a nominal max ground normal deviation (global, deviation from the ground
effect plane).
Lighting system needs to self-regulate the number of pixels it draws
(pre-emptively please), by shrinking the WS bounds of lights that are farther
away but big on screen.
Sniper scopes though, they
were always going to be a bit of a problem.Not going to think about them till I have to.I'd like to include them, but not if
they're going to be more trouble than they're worth.
Light ranges are now read in
from the NWF and applied to projectors and to DeferredLighting.fx (via
texcoords).
Lost a heap of time today
because of the shouting.I've worked
out that it's someone next door screaming at the football, which is a big
relief.The sound of someone picking
fights at the top of their voice is not a good sound.
If I can work out which team(s) are causing the problem, maybe I can plan
around it.Or buy him some tickets so
he can do his yelling in the stadium.
Terminology: I'm using Render (rather than Draw) for more
'final' rendering (even if it's still off-screen), eg. CLight::Render.
Using Generate...
for intermediate off-screen rendering, eg. GenerateDepthImages,
GenerateShadows, GenerateImpactImage, GenerateReflectionImage.
Added CLight; dispersed
CProjector::DrawAll into CLightManager::Render, CLight::Render and
CProjector::GenerateShadows.
9-Mar-2007
0
Night off (knackered from last
night).Curry and film.
Film due back Fri 16th.
This weekend I'm working on
puddles and ripples
8-Mar-2007
7.5
Options: Replaced
'general\aspectRatio' with 'realWorld\displayAspectX' and
'realWorld\displayAspectY'.
Checked that these work properly (checked shape of spheres).
g_game.aspectRatio now replaced
by g_game.options.realWorld.viewportAspectRatio
Stereoscopy no longer disables
HDR bloom & glares.I don't know
what I was thinking there - the post effects actually make the stereo image
much *less* confusing at the moment.
3DTV: Paint Shop Pro, Windows
slideshow etc, they all fail to give proper pixel-to-pixel full-screen views
of images - they add nasty little borders.These borders stop the TV switching into 3D mode because the header
pixels land at the wrong coordinates >:(
42" WOWvx TV: display
width: 93.26 cm based on 107cm for the 'diameter'.
Options: End-of-line '//'
comments work fine.Maybe support full
// and /* */ coments later, although when the menus go in you should never
have to edit the text files manually, unless something goes wrong.
Checked that my 3DTV colour and
depth sides are matching up pixel-to-pixel.
Added stereoscopy options
_3DTV_nearDist
_3DTV_farDist
_3DTV_disparityPower
_3DTV_headerTexture
Checked that these all work properly.
3DTV: Had a load of trouble with
header pixels not coming through at exactly their original colours.Turned out to be D3DXCreateTextureFromFile;
found this explanation in the DX help:
"Filtering is automatically applied to a texture created using this
method. The filtering is equivalent to D3DX_FILTER_TRIANGLE |
D3DX_FILTER_DITHER in D3DX_FILTER."
Finally got the 3DTV mode
working.Just
crossing my fingers that it'll *actually* work tomorrow.<-It does - YAY!
#version 10.4
Released v10.4.This has the 3DTV mode in it.
7-Mar-2007
4.5
Been clearing a path for the
3DTV mode which will be in a 10.3 build very shortly.Improved and rearranged various code for
this (options, stereoscopy, hdr).
3DTV mode currently
forces:
fullScreenXRes:1920
fullScreenYRes:1080
frameRT:[/2, /2]
Need to remember to omit the header pixels when windowed.
v10.2 has tone-mapping blackouts
on nVidia 6800 & 7800.Moving the
mouse and alt-entereing a few times gets you past this, till the problem area
(whatever it is) comes back into view.There must be a divide-by-zero in a pixel shader somewhere.Anything to do with the sky?Should go through all the shaders at work
fixing potential db0's.Would be nice
if the DX software emulation could trap these for you somehow?
NOTE: g_game.GetDisplayWidth
returns the display width not the current RT width.However, this is also the viewport width, regardless of the
RT dimensions (this is evident in CRainImpact::GenerateImpactImage).Same goes for GetDisplayHeight.Maybe rename these?
<- Doesn't seem to be the case now; see 22.5.7
Options: ReadBool now accepts
t/f, y/n, v/f, o/n, yes/no, oui/non, vrai/faux.Only the first letter is checked.Not case-sensitive.
Options: Added 'HDR\middleGrey'
and 'HDR\finalMultiplier'.The latter
is especially useful if tone-mapping is disabled (set it to 100 odd).NOTE: applies before bloom and glare,
doesn't multiply those.
Remember '\n' works in
UtilFatalError etc, all the message boxes.
Tone-mapping now pauses when the
game is paused.
6-Mar-2007
3.333
Fixed the anaglyph
stereoscopy.POM looks great in
stereo, although at the same time its intrinsic artefacts become more
obvious.
Would love to get POM pushing outwards rather than inwards, that would be
'the future'.Need to look into that.
Rain splashes look ok in stereo,
I had nothing to worry about.The
shapes in each eye don't match but the positions do which is the important
bit.
#version 10.2
Released v10.2.Stereoscopy working again.
Custom.txt profile is now 'permanent' meaning that the player's
customisations persist between versions, yay! (provided the option/category
names don't change).
Default.txt profile is now read-only because it should never be edited.
5-Mar-2007
4.83333
SVN: Got into a real tangle
after using 'update item to revision' instead of 'revert changes from
revision' :/
Darned prepositions.
Finished the splash
blending.I'll leave it to soak now,
ie. get on with the rest of the rain effects and come back to it later to fix
anything that bugs me about it.
#version 10.0
Released v10.0.Stereoscopy is unavailable in this build.
4-Mar-2007
8.5
Just noticed the RenderMonkey
'FireBall' sample (Fire.rfx), which is really quite tasty and very
simple.Might have a look at that
again when I'm ready to do explosions.
(below) Rain splash blending
going well, nearly finished.Looks
really nice pitter-pattering and catching the lights - adds loads :)Doesn't lend itself well to being screen-grabbed
though.
3-Mar-2007
11
Got some decent footage last
night, including puddle videos.The
rain still wasn't as heavy as I'd like though, but better than last
time.Last time it was 'light', this
time it was 'moderate'.I want 'heavy'
or 'severe'.The waterproofs were a
good idea.
#version 9.1
Uploaded private test v9.1 for
M.This disables HDR multisampling -
just in case that could have been involved with his zbuffer problems.
Noticed that the
(SetRenderTarget, Clear) pattern is used in lots of DX samples, so I assume
it's the right way to do it (rather than using ColorFill).
ProjectorShadow.fx: Tiny shadow
improvement to stop landscape getting a dark inner fringe when set against
the sky.It was an early 'clip' so
should serve as a good optimisation too (er, when you're looking at the sky
;)
Had an embarrassing bug with a
for loop.A semi-colon and a comma had
ended up swapped, so that a variable was getting re-initalised as part of the
condition each time round.D'oh!
:|Would've been easier to spot if I
hadn't been faffing with FVF codes / vertex declarations at the time.
#revision
363
Animated the rain splashes, also
improved their shape a fair bit.That's the first time I've EVER used point sprites (for the impact
dots), but they were definitely the tool for the job.
There's currently only 48 dots, covering a 3x3m tile.Render target is 512 like the stand-in
texture was.
TODO:
- maybe use a displacement map (some tesselating
pattern) to shuffle up the planar mapping of the impact dots over the
environment.You can see the tiling at
the moment, but maybe the lighting/blending will hide that.
- noise normals need to account for viewport
dimensions, aspect ratio, etc.
- need to prevent big tall impact streaks appearing down the walls (tricky
because of the normal mapping)
- maybe try scrolling the noise normals down to
make the droplets fall?Or upwards to
make them jump?Or maybe one of
each?Just so you don't see droplets
hanging in the air.
So, onto the
lighting/compositing of these splashes then...
REMEMBER: Non-enumerated FXOs
are not yet reloaded by Reload!!That confused me for a while, tweaking
SolidColour.
Tweaked the lighting again to
get the place looking dark & wet for the splash compositing.
#revision
365
Fixed a bug where HDR
glares/bloom were going missing on Reload (needed to set some effect
variables in the restore).
2-Mar-2007
4
The errors yesterday were
because I had re-enabled multisampling in the HDR system at the last minute
before building v8.1 I think.RT and
D/S surfaces have to match in terms of multisampling.My RTs don't use multisampling so they were
conflicting with the D/S buffer set by the HDR code.In order for me to get any benefit out of
FSAA, I think I'd need to enable multisampling for all four MRTs used for the
main rendering.I get the impression
they would need to be created using CreateRenderTarget not CreateTexture, and
to use them as a texture I'd have to StretchRect them onto one.
I might come back and put FSAA control in later.I don't want to waste any time with it now;
I've disabled it in the HDR code again.Stuff like motion blur & DOF should greatly reduce the need for
FSAA anyway...
SVN: What on earth does
'incomplete' mean? (Noctambule & Source folders).Reverting doesn't fix it, nor does
cleaning.
Added the CRainImpact class
which updates and renders the rain impact effects.Belongs to the Environment.
There's proper rain out
tonight.Hopefully more where that
came from.TODO: flash photography of
impacts on various surfaces (puddle/car/road/pavement/brick) in the same
heavy shower, for splash shape/animation tweaks.
Considered breaking away from
the project's prescribed coding style to experiment with C++ rectitude and
develop more thoughtful conventions.Then chickened out.It's not
that I don't want to learn the most correct ways of doing things.It's just that I can't afford to lose any
time being indecisive about code, and since I have no idea what exact style
would best please the target audience anyway, I'd rather give them the
impression that I can stick to whatever style guide I'm given.By all means if there were to be an Exhibit
B, it would be good to use a contrasting coding style for that...
1-Mar-2007
4
Slightly worried about
stereoscopy for these rain splashes.Currently they'd be very inconsistent between the two eyes (can't test
at the moment).I'll wait & see
how they look in stereo once they're lit & animated.
Spent more time faffing with
device handling stuff, and got annoyed.Still isn't completely clear.Alt-enter does not destroy the device like the docs suggest it would,
so the destroyed-device situation has never been tested and would probably
crash?
Ordering is a big part of the problem too.The code implies assumptions about when things are going to happen to
the device in relation to when things are going to happen to class
instances.Every time I deal with this
stuff I get annoyed, because it doesn't feel like it progresses the game at
all, just feels like a horrible waste of time.
InitDeviceObjects /
OnCreateDevice: (The device is ready)
- Ready to create managed resources, ie those in D3DPOOL_MANAGED.
InvalidateDeviceObjects / OnLostDevice: (Device has become
unavailable)
- Get rid of pointers to tetxures, effects and render targets, except those
in D3DPOOL_MANAGED.
RestoreDeviceObjects / OnResetDevice: (Device has come back again)
- Get pointers to textures, effects and render targets, except those in
D3DPOOL_MANAGED.
DeleteDeviceObjects / OnDestroyDevice:
- Must now delete managed resources, ie those in D3DPOOL_MANAGED..
Tried changing CRenderTarget
over to use D3DPOOL_MANAGED but the D3DXCreateTexture failed with
D3DERR_INVALIDCALL.No idea why, it
didn't tell me.
At least now the zombie material
is set up properly so it supports tweaks while the game is running.
Spent the whole evening doing
faffy housekeeping stuff :(
TODO: Fix all the Debug-D3D
errors that I'm now getting (when did this start?)
eg. On DEVICE->Clear.This kind
of stuff:
"Direct3D9: (ERROR) :MultiSampleType between DepthStencil Buffer and
RenderTarget must match."
Could this have anything to do with M's v8.1 glitches? <-v9.2 will
determine this
28-Feb-2007
2.75
TODO: blow these splash
extrusions around according to wind - that will be lovely.Bonus points for a 'wave' effect when it
gusts.
Experimented with multi-pronged
splashes but it was too hard to disguise what was going on.Instead I fattened up the dots on the
placeholder texture and broke up the splash tips with a noisy normal map.
#revision
351
(below) Pretty happy with the
splash shape now, so I'm going to move on.Tossed a coin as to whether to animate or light it next - animation it
is!
27-Feb-2007
4
Got the rain splashes projecting
out away from the surface normals.Tried having them reflected by the normals but it looked cack.It made the problems more obvious, made the
splashes twice as bendy which I don't want.
Artificially 'flattening' the splash direction seems to work better,
keeping the length the same (ie. making it less vertical)
From now on I'm making sure that
normals are always normalised before being written to the normals
buffer.Saves me having to worry about
it in the deferred shaders.
#revision
346?
(above) That was a bit flukey,
the breakup is caused by the ground normals.Shows promise though.
TODO: Try to recreate that shape in a more controlled way (use a fine
random-noise displacement/strength map, affect far ends of the splashes
more).<-done
26-Feb-2007
2.83333
tex2Dgrad is a nice cheap way of
doing anisotropic blurs.Could be
useful for the motion blur.Or for the
falling rain even?
D3DXComputeNormalMap, eh?
#revision
341
(below) This is *just* using
tex2Dgrad with dy*30 - I'm not doing any of the work there, tex2Dgrad
is!
Those are single texture samples!!
I like tex2Dgrad.It's good.
EXCEL: Can't ctrl-V images into
the sheet, if you do they export as giant EMZ files even after you've told
Excel to compress them & reduce resolution.
EXCEL: Need to remember to tell
it to compress all images.Use
"web/screen", "compress pictures" and "delete
cropped areas".
#revision
342
(below) Early test with a
(literally) v-shaped filter.You can
see where I'm going with it.
25-Feb-2007
10.998
WEBSITE: added subdomains
'devlog' and 'lecauchemar'.
Devlog file is now devlog/index.htm
Automated devlog image file
handling.There's now just a single
'devlog' folder for me to copy up each time, and I can (hopefully) skip the
overwriting of the image files, unless I've gone back and changed/resized old
pictures or that.
Added a NoctambuleShaders
project by copying the main project and deleting everything but the shaders.
FXOs are now built to
source\debug, source\release and source\master, depending on config.This is tons better.
Tidied-up effect loading.Doesn't do any searching now so should be
faster.
Rain impacts not extruding in
Master - what's that all about?<-(26th) was using an
increment variable in a for loop that did texture samples.Which is something that doesn't seem to
compile properly.Had the same problem
with ShadowBlur.fx on the 27th Jan.I
haven't looked at the disassembly yet; the important thing is to remember to
avoid doing it.
NoctambuleShaders.vcproj: Needed
to set project type to 'exe' otherwise $TargetDir was empty (so FX build rule
compiled to the wrong place).
TODO: support reload of zombie
material (CAnimMesh needs an InvalidateDeviceObjects)
Currently that material is skipped in the reload<-FIXED
Made a Shaders.sln because it
didn't seem to be possible to compile FX files while the game was running,
unless they were moved to a separate solution.This is working now :)Seemed to cause a couple fo GPU
hangs/resets but now seems to be behaving again...
Shaders.sln contains all the FX files (set up to build exactly as before),
plus a ShaderBuild.vcproj which, when run, builds every FXO in every config.
Added a CNoctambuleApp::Reload
method which is called on gain-focus and on spacebar.Currently re-loads all the enumerated
shader effects.Will add re-loading of
leam.x and leam.nwf later.TODO: add an option ('debug' category) to prevent HDR re-init
(this causes a gentle tonemapping burst which could be awkward when debugging
something visual).<- sorted, see
g_game.flags.fakeDeviceLossForReload.
24-Feb-2007
12
TODO: Support Max 'groups'.Currently, grouped scenics don't show up
in-game.
Try to remember that ZENABLE
means 'Z TEST ENABLE'.I wish they'd named that properly.
Fixed the reflections of cars
and zombies cutting out - vertexcolour.fx wasn't zwriting/ztesting so
reflected skydome was covering them up.
MAX: Tesselate modifier is tons
neater than Subdivide.Now using it on
the cars.
CONTROLS IDEA: hold space to
crouch; tap space to jump?
#revision
317
Made a really important fix in
Mesh.cpp: compensation for normals that are crushed by irregular object scale (I had kind-of imagined Max would fix this for me but it
doesn't).
Cars' lighting was very broken this morning but now it looks lovely :)
Had to reset the scale/transform
on generalLandscape, otherwise rotating the test objects made them shear.
TODO: move projector::drawall
into environment::rendersubset<-done
Spent an embarrassing amount of
time trying to work out why my rain splashes weren't filtering.Turns out UtilDrawScreenQuad was forcing
the filtering to POINT.That'll do it.
CHECK THE OBVIOUS!!!
Used the CTexture class (for the
first time) for the rain-splash dummy texture.Nice & easy to use.
Rain splashes going very
nicely.TODO: use
tex2Dgrad to counteract the ellipsing of the impact dots?Can maybe then switch back to lower-res
(fixed-point, filterable) source RT.<-done and
done, works a treat.Had to turn off
mipping now I'm biasing the lod though.
Realised I could add pictures to
this Excel doc.Pictures in a dev log
are good - I'll probably automate the file handling of this tomorrow.Here's some pics of my placeholder cars and
the beginnings of my rain splashes.The two spheres are Wet Ball and Dry Ball; they show the min and max
gloss levels.
23-Feb-2007
4.75
#revision
314
That revision is a safe copy of
the early parallax rain impact test.
I'm not going to use that method because although it can look great, it
really works by fluke and could therefore be difficult to tame.
I've gathered some valuable info from resurrecting it though, which I
intend to use in the proper version.
Eg. using the splashes/droplets as a screen-aligned height map for
refracting the lighting/scene.The
gist of the effect was really just those cute blobs & plips refracting
the radial ripple highlights behind them.
Came up with a plan for rain
impact splashes.Found good reference
videos at www.lucidmovement.com
Added some placeholder cars to
help test the rain effects.Gave them
the dimensions of an Audi TT because they kinda turned out that shape.
22-Feb-2007
3.666
Rain impacts going fine. TODO: consider sorta v-shaped screen-space splash filter.Use a downwards-offset sample of the
position buffer to determine the distance of the splashing surface, for
perspective.<-doing this stuff now
21-Feb-2007
4.1666
Learned about SVN tags and
branches.I'll create a branch for
every version of the game.
Branched v9.0 and made a couple
of tiny fixes (missing files, literal path).9.0 gives M exactly the same bug as 8.1 did.9.0 and 8.1 both work fine for me at work
though.
I'm not going to investigate this till after the rain milestone.It's not like it's crashing or anything,
and the rain is URGENT.
SVN: MERGE: Select Merge from
the DESTINATION folder, not the source.
MAX: Can't save DX shader
materials to a material library - it doesn't remember the textures and uses
the FX's defaults instead.
Also If you try to explicitly load a library containing a DX material
(rather than just putting it to the default 3dsmax.mat), you get this
crash:
Error loading ParamBlock2 / Invalid file - only partially loaded
So I'm just going to have to
'make do' without being able to safely store my materials.Anything I especially want to keep, I'll
just have to map it onto a non-exporting dummy object >:/
It's bad, I've already lost some materials.
Got some slightly more
encouraging results with the rain impacts...
20-Feb-2007
3.25
TODO: REALLY NEED TO IMPROVE
SHADOW CULLING.Many objects' shadows
go out of view when the object is out of view, but currently the shadow is
always still being generated in this case.
#version 9.0
Released test version 9.0 en to
the M.Not public.
v8.1 exhibited zbuffer problems
chez M.Made some code changes in the
hope of fixing this (v9.0)
19-Feb-2007
2
Did a general reference shoot in
light rain this evening (properly dark).
Well worth doing, lots of useful stuff to keep me going till it rains
properly.No real rain effects this
time, but plenty of wet surfaces and lighting/shadow reference.Plenty of nice water on cars, but my camera
wasn't good enough to capture it clearly :(
Cling-film for waterproofing worked ok.
Corrected my registrant details
(at nominet.org.uk) for all domain names except the .org (which they don't
deal with).
TODO: have a think about light
that bounces off windows and then casts out to light the scene.A few of my photos from this evening show
this very clearly.Could be one of
those effects that adds a suprising layer of realism, and my gut feeling is
that it's very doable.
Made a nice de-oranging
make-it-look-the-way-it-actually-looked filter in CS2.Fixes my photos nicely (testadjust.psd).
18-Feb-2007
10.2497
TODO: parallax.fx: deal with
g_vTextureDims - this can be neatly driven by the 'texture res aspect ratio'
property.
POM shearing was caused by
irregular U/V scale
Could HDR blackouts have been
anything to do with those luminous CRUK windows?Now hidden so they don't export.
Backface culling is definitely
the wrong way round on PZ shadows.<-FIXED, TODO:
neaten/investigate.
Still doesn't allow me to get PROJECTORSHADOW_DEPTH_BIAS down to ~zero.
POM: To prevent shearing,
textures must always be mapped with REGULAR SCALE; any irregular scale needed
(to match the texture's natural world-space aspect ratio) should be effected
using the material's 'WS Texture Width' & 'WS Texture Height' properties.
Can easily add a Width Factor / Height Factor
later if it's useful, but I wouldn't want to add it without good reason.
NOTE: 'Scale' parameter in
nVidia normal map filter is height map scale in texels.
My system is now such that normal maps must
always be exported with the 'height scale' set to equal the texture x
resolution.ALWAYS ALWAYS!
Also remember to use the 'wrap' option.
Photoshop: Converting from
Greyscale to RGB Colour degrades the image unless you switch to
16bits/channel FIRST.
TODO: Make a photoshop action to convert from a height
map right the way through to a normal map, via the optimal route.<-DONE(20.4.7)
Can't find 'Noctambule
Actions.atn' (Photoshop actions) must've left it at work.->nah
can't find it there either, ah well, it didn't have much in it, just a few
tests really.
NOTE: there is no problem with
converting non-square height maps into normal maps, as long as the texture
ends up mapped at its natural world-space aspect ratio (as it should do with
the system I've put in place).
TODO: Max_Parallax.fx: free up
some arithmetic slots so that PSProcessTangentSpaceNormal can be called.
Remember when testing specular
POM results that the deferred lighting positions don't yet take into account
the POM (need to sort out shadow stuff for that to work).
Colour map alpha is now gloss
level.This is multiplied by the
parallax material's 'gloss range' property.
TODO: Might want to map 'illumination colour' to control bumpiness at some
point?
TODO: Set up a Max material
library so nothing gets lost.
MAX: when you create a new mesh,
you have to 'initalise' its vert colours by tweaking them a bit, otherwise
they'll be random nonsense in the game.
Non-mesh objects will have uninitialised vert colours in game :/
Parallax material now shows
vertex alpha and vertex gloss, both in Max and in game.
TODO: What scale should my UVW
modifiers be using???
<-UVW modifiers should use a width/height/depth of
PARALLAX_REFERENCE_TEXTURE_DIMENSION, which is set to the default dimension
for the UVW modifier, 1 metre.
Spent most of the day getting my
normal+height map system nailed.Extremely pleased with it - the normals ending up in the WS normal
buffer are now the ACTUAL, CORRECT normals for the surface, based on the POM
extrusion, the UV scale, and the natural world-space dimensions of the
texture.Getting the normals exactly
correct is far more important when using POM than it is when using 'bump
mapping' or squidgy 'parallax mapping'.
TODO: chuck together a no-frills document explaining the mapping system
(partly so I don't forget it all), and put it on the site.Might be useful for people at work.
17-Feb-2007
11.834
Now using the prefix 'c' for
shader constants.
Some of the constants in any shaders I export from RM will be like
'__SpecularPower' etc, that's so the RM FX exporter knows which semantic to
give them.
MAX: Why on earth do UVW
modifiers sometimes inherit a scale that the object doesn't have??!
- tried resetting object's transform & scale before adding the
modifier
- tried converting to editable mesh before adding the modifier
Noticed that I had a typo that
was saying vertical fov = aspect ratio each frame!Fixed.
Fixed auto-fov.Gives a much lower fov than fps players
will be used to (in fact they'll probably hate it), but I think it brings you
more into the scene, as if it were real.I also think a low fov is particularly well suited to a dream
environment because...well that's the way I dream - a bit tunnel-visioned,
more concerned with details than periphery, never knowing what's round the
next corner.
Plus it will balance better with the degree of detail that I want to show
off in this game.And it *totally*
justifies the POM.Placeholder
brickwork is currently set to 1cm, yet the benefit is strikingly visible.
Got a GPU hang in fullscreen
mode after fixing the fov (POM not benefitting from mipping as much,
etc).Used the reset button.When I came back to it it was in the dos
startup saying 'DISC BOOT ERROR, PLEASE INSERT SYSTEM DISC'.Powering off and on again fixed this.
Reduced the number of samples in
the POM, seems happier now.Still
getting up above 90°C (fullscreen) though, with no overclocking at all.
Added some useful properties to
the parallax material.These
future-proof against changes to textures (of which I'm sure there'll be
many).TODO: keep these in a header
for all landscape materials to use:
Natural world-space measurements:
- texture width
- texture height
- texture depth (height map distance range)
Other measurements:
- texture res aspect ratio
Adjustments:
- depthFactor
MAX: Can't choose DDS files for
dx shader materials?!
Colour-burned sharpen for
wetting effect?
TODO: support the
PROJECTOR_ENABLED define being 0 (currently gives black screen)
TODO: support tone mapping being
disabled (currently super-dark)
Found that a lot of my landscape
tangents & binormals were knackered, hence the POM faults and the faults
I used to see with non-deferred lighting.
Tried exporting leam.x with 'optimize mesh' set to normal / none - made no
difference.Usually set to
'optimized'.
The ATi POM demo loads all surface vectors from
the model file, which is in some format of their own.
<- FIXED.Just had to pass true for the
bSplitVertexForOptimalTangents parameter of CMesh::SetVertexDecl.
Trees pom is still weird even though their surface vectors were fine all
along (only 'square' stuff had bad ones).
TODO: investigate
I'm going to carry on using
subdivide modifiers, but only very MILD (coarse) ones.There are some genuinely bad polys in the
landscape.
TODO: couple of Max
scripts:
- one that exports leam.x & leam.nwf
- one that adds the selection to 'buildings' layer and connects it to
'generalLandscape'
TODO: bumpiness (normal)
control?Tie this into DepthFactor?
Sketch in diary in 28th May
shows vertex shader requirement for protruding POM.
Still got some minor shearing on
my pom bricks.TODO: Get to the bottom
of this!!!!<-fixed see 18th
Made some improvements to the
lighting model, including a Fresnel term for the specular.The power for this is dependent on the
gloss level (less glossy = higher power = more "reluctant" to show
specular highglights).
16-Feb-2007
2.41667
PAYPAL: Withdrawal (transfer to
bank) charges: Free over £50, else 25p
Noticed there is a 'MySpace
Film' network now.Registered the
addresses 'lecauchemar' and 'programmerart', just in case the site is stable
enough to be worth using in a year or two.I hate MySpace.
Login details are in the 'archive' folder for now - need a neater place to
file all correspondence later on.
FX and FXO files still live in
Source\HLSL.FXO location isn't set in
stone, makes no difference to the build process after all.But I don't want the
FX location to ever change:
IMPORTANT: Moving the FX files resets all the
material properties in leam.max!!
NOTE: next dxsdk install should
automatically install the plugins for CS2, I'm hoping.Till then I'll use Photoshop 6 for likes of
normal-map generation.TODO: maybe
sorta 'hide' photoshop 6 when I install dxsdk?
Still playing with the parallax
shaders and planning the rain impacts....
15-Feb-2007
3.166
Got the shadows back into a
decent state.I've made all the depth
images backfaces-only, which makes the shadows connect properly with their
casters (was previously a big nasty gap).
For some reason PZ shadows still don't connect
properly - TODO: FIX THIS!<-FIXED, TODO:
neaten/investigate
Wasn't able to reduce PROJECTORSHADOW_DEPTH_BIAS much (glitches, not sure
why), but it doesn't seem crucial to do so anyway.
Had another bash at getting the
FX files building neatly, still couldn't do it.I'm not sure if I'm wasting time by even
trying to make it work - might just not be possible full stop.
Found that I can compile a multi-selection of FX files though - that might
be helpful in some situations.
Had a look at the DX samples but they cheat totally by not setting any
build tool on the FX files.
...Eventually managed to make
*some* progress on the FX build situation.It now knows which FX files have changed and only compiles those.Unfortunately it still only compiles one
file per build, but it's still better than it was.
Found that skipping the assembly txt output speeds up the compile a lot.
TODO: finalise FX & FXO locations; get the
game code to find the FXOs in the right place!->done
see 16th
Removed
CLandscape::RenderGroundDepthMask into stuff.cpp
14-Feb-2007
4.25
Fixed one of the shadow problems
(h_Projector.fx \ ProjectorDepth_PixelShader, DEPTH output was being
re-divided by the projector distance range).This fixed the new(?) glitches on east lamppost and west tree.
#revision 272
TODO: FIX: Noticed that zombies
are not receiving landscape shadows!!<-FIXED
Rearranged the HDR stuff into
class C_HDR (hdr.cpp/h).Tidied out
junk.Moved the magic to be around the
environment draw, not the entire game draw.
TODO:
- Get the HUD back, it's become hidden now
- move stereoscopy to be inside the HDR magic (around env draw) rather than
around game draw
Realised that only the BACKFACES
of objects (need to / should) write to the depth images.This allows a smaller (0?)
PROJECTORSHADOW_DEPTH_BIAS (^_^).
TODO: GET THIS WORKING!!!<-see 15th
If doing self-shadowing POM, need to draw the front faces too, but only the
backfacing pixels thereof.
13-Feb-2007
3.75
Can't get any blackouts now (at
home) what's going on?Removed the
'saturate' in deferredlightingapply
TODO: Test for blackouts at work
Found that tweaking the HDR 'key
value' is probably the best way of brightening/darkening the scene.TODO: put this into an
option.<-done
Started planning raindrop
impacts, got some photo reference...........
Ported the DX parallax mapping
and parallax occlusion mapping shaders; got both effects working nicely on my
placeholder brickwork.The POM is
adding a lot, makes the walls feel chunky and tactile.Can't wait to use it on proper textures.
Adding a temporary subdivide modifier to
bedford court cleaned up the results a lot, but there's still some odd polys.<-FIXED see 17th.
12-Feb-2007
2
Sent out the first news
letter.Took ages, which is one of the
reasons why I won't be making them very frequent.Not sure how to log the time spent on stuff
like that (or on the website for example).I'll log it like it is for now - it is time spent on the project after
all, kinda.Maybe improving my pace on
that kind of thing is something I can work on.
Don't use HTML next time, too many glitches.
Added a 'Newsletters' folder to
archive the news letters.
The v8.1 download works just
fine on the 7800 at work, phew.
11-Feb-2007
5.66667
Got my charity account working
in PayPal finally, set up a donation button on the site.Doesn't accept payments from me (error
3005), but hopefully it works for anyone else(?)
NOTE: do not set up bank funding for (or "confirm") a savings
account!!That's 'crossing the
streams', apparently.
Noticed that puddles are
changing size/shape depending on viewing distance.What's that all about?Mip-mapping?Fogging?
TEMP BODGE: Added a 'saturate'
at the end of DeferredLightingApply - this completely prevents blackouts, and
doesn't currently look bad.<-removed now
Got the test for floating-point
blending working - FOR THE MACHINE AT WORK.On my ATi the test query result says it can do float blending (in all
device configurations) but it can't.
So I'm now just using 2TTT unless the query says that the card can't blend
to it (options.cpp).
Tone mapping is very
unresponsive at work - like, it barely changes.Much nicer on mine for some reason.
TODO: get rid of the 'UI'
folder!
Noticed some of the shadows are
suddenly worse than before (eg. east lamppost, west tree)->FIXED
#revision 272
#version 8
Released version 8.labelled the commit "version 8"
4-Feb-2007
11.833
Cyan & green look
practically the same through the red filter, at any brightness.Green leaks into red tons more than blue
does.In fact red blocking blue is the
only filtration any of these glasses manages to get right :(
Got deghosting working in all
the anaglyph combinations (using defines), although it currently forces cyan
to blue.
Added some more example
profiles.
7800: seems I can create 2TTT
targets just fine but not blend to them.
LCD monitor at work filters
slightly more cleanly than my CRT at home.In particular the green & blue filter out very well (blue looks
just like black).
failed an assert one time when
starting up the master build: assert(cameraEye==centre)
got a few blackouts.At work they don't fade back up so you have
to alt-enter which fixes the blackout instantly.
7800 does do float blending,
phew.
Caps viewer reports 7800's
memory as video:256, texture:512
more parts of the image make it
into the HDR high pass for some reason (eg. diffuse patch on bedford
court).Worrying.
managed to remove several
zbuffer clears.Added a new one into
projector::drawall (needed to otherwise frame size < screen size caused
zbuffer junk)
remmed out 4
setrendertarget(0,NULL)'s because this is illegal (they were just crazy
random tests anyway)
Can't seem to prevent blackouts
by clamping luminance any more, hmm
Blackouts are getting really
annoying, but easily reproducable at least.Shouldn't really put a build out till I've fixed them.See note above,that's odd.More tinkering required :/
Might be a divide-by-zero in a pixel shader somewhere...
Fixed the crazy feedback in the
ground reflections (looked like the world's lamest fire effect and was
therefore a show-stopper).
Had just been using the ordinary deferredlightingapply rather than the
reflected one.
This green fill colour means
I'm at work using an nVidia 7800.
Couldn't manage to test for
floating-point blending support...Asked on the work forums...see if anything comes back.
CheckDeviceFormat with D3DUSAGE_QUERY_POSTPIXELSHADER_BLENDING always
returned D3DERR_NOTAVAILABLE, no matter what format I checked
Tweaked the lighting so it
looked decent on the 7800, no idea what it'll look like at home...
Failed to put version 8 out the
door :(Won't be able to do any more
for another week (Leeds gig).
3-Feb-2007
6.83333
Found the wxdebug 'REMIND' macro
and adapted it to suit.Eg:
#pragma REMIND("temp test")
outputs
c:\noctambule\source\options.cpp(201): REMIND: temp test
This game, formerly codenamed
Project 'Noctambule', now has a name:
Le CAUCHEMAR
Checked with Mr. B that this didn't sound
totally naff or anything.It's quite
scary choosing a name for something as important as this, especially a name
in a foreign language.
The letters have a nice shape to them, I'll enjoy doing the titles.
I've only heard of one film with this exact name, and it was made in like
1893 or something!I'll have to watch
it some time.
Bought lecauchemar.com from the
usual place.Got it for 3 years.All the other domains are free as well,
surprisingly.
lecauchemar.com currently points to programmerart.org/lecauchemar.
All mail ^lecauchemar.com goes to ^programmerart.org
I'm almost hitting my VRAM limit
:/Was crashing when I upped the
patchwork colour RT to screen-size.
Need to sort this out...
Got rid of the 'FRAME' RT in
monoscopic mode.
TODO: Can do the same for anaglyph but not when
deghosting is enabled.<-done
Merged AnaglyphFrameCopy into
DeferredLightingApply, using an ANAGLYPH define.
Added an ambience component into
the lighting - looks good.
Fixed the blockiness that had
started with the shadows (needed to clear the source RT to full 'diffusion'
(green).
Wasn't on great form today, need
to pick up the pace tomorrow to get that build out.
2-Feb-2007
2.5
Got the game compiling and
running at work (didn't need to do anything except make sure all me files
were in SVN).
Was pleased to see the reflected lighting corruption is fixed now.
DXUT: command line: use eg.
"width:320 height:240", not '='.
Added a 'Custom.txt' profile
that ships and always overrides Default.txt.The player is advised to edit Custom rather than Default.Also added some example profiles for the
player.
Got the "fullScreen",
"fullScreenXRes" & "fullScreenYRes" options
working.
Also added "windowXRes" and "windowYRes" which are
working nicely.
BUG: Mouse pointer appears in
full-screen mode when res is extremely low (like, 320*240)
The game looks like it's well on
target for a version 8 release this weekend, probably early Sunday evening.
TODO for version 8:
- Compatibility work on nVidia 7800 (lighting fixes)
- Optimise lighting for cards with floating-point blending?Still haven't found the caps flag for that
yet.
- Spruce the place up a bit.Put new bulbs in the streetlights.Build some ground detail, don't be shy!
(pull textures from Photography\Texture\TextureShoot_17_9_5).This process should tell me whether or not
I need to buy a new camera (probably should).Do that 'texture dimensions in the VS' thing??
- Improve & optimise PatchworkDimpleProcessing.fx?
- Fix remaining anaglyph modes and get ghost
reduction working for all of them.
- Give the project its final title (and buy the
domain)?
- DON'T try to make any shadow fixes.Lighting tweaks are ok though.
- Was thinking about putting some parallax mapping in for the buildings.
1-Feb-2007
4.16667
Pragma'd-away some warnings
(Global.h).Using push&pop for
constant conditional expressions (at least trying to - the compiler seems to
be ignoring me (?)).
Not totally clear on C4238, which I've disabled for now ("nonstandard
extension used : class rvalue used as lvalue").
Noticed that DX HDR sample has a
flash of over-exposure whenever you resize its window or
alt-tab/alt-enter.Also just after
alt-entering / alt-tabbing, you see a one-frame flicker like I get on mine
(no visible corruption in the flicker though).
Options are now re-applied when
the game regains its device (so window size changes / alt-enters work
properly).
DX HDR sample, when set to
startup full-screen, also does the initial 640*480 black box thing that I get
on mine.I get the feeling it's
harmless.<-Nah it doesn't if done by the command line,
hmm
NOTE: Apparently if you use the
desktop resolution for fullscreen it avoids a slow mode change.
Nearly got the
"fullScreen", "fullScreenXRes" &
"fullScreenYRes" options working.
Added "language"
option.Need to work out a better way
of doing the two different installs though.Having two totally separate install projects is crazy (keeping them
both up to date as I add files).The
only file that needs to be different now is the Default.txt profile.
31-Jan-2007
2
Distorted ground reflection now
draws to its own float RT which is used as a texture in
DeferredLightingApply.
So good to get rid of that banding I had before - phew!Gets rid of a nasty buffer copy too.
Started using a nice easy method
for FX variations, eg. reflected & non-reflected.The reflected version is just two lines
that say
#define REFLECTED
#include "TheEffect.fx"
Job done!No faffing with
headers.Should've thought of this
earlier.
Having the enemies only visible
in reflections & shadows looks quite scary, hmm...
TODO: should be able to use the
ground reflection (source) image as an environment map for weapons etc, if
I'm careful about it.
TODO: work out why the reflected
landscape render is covering-up the reflected zombies (was the same yesterday
I'm pretty sure.
<-(1.2.7) Could not work this one out - used
a horrible bodge for now (enemies are placeholder anyway; their render will
change)
"pMesh->DrawSubset" (for the landscape) was the line that was
causing the zombies to get hidden - exactly as if the zombies hadn't written
to the zbuffer (which they do, of course)
30-Jan-2007
2.666
Looks like the reflections are
lagging 1 frame behind?
The 2TTT specular buffer is not
up to the job of holding the ground reflections and specular on the same
scale - the banding is fatal.
So here's the plan:
TODO: Make a new float RT for the ground reflections
to render into (needs to MRT-in with diffuse+position+normals).After the deferred lighting has applied
over the scene, add the floating-point reflections, which are already
weighted by ground gloss.Remember to
use the nice clean version for cards with float blending.<-done
29-Jan-2007
2.83333
Got everything sorting properly
again.Right mess at the moment
though, needs a cleanup.
TODO: TIDY UP THE ENVIRONMENT RENDER!
TODO: GET REFLECTED LIGHTING BACK
Quickly tried 'wetting' my
placeholder tarmac-type textures.Instantly gave the effect I needed, making the road paint stand out as
it should.
The new lighting model is lovely, a zillion times better than what was
standing in for it.
She's actually looking quite pretty at the moment; version 8 should be
quite the looker :)
I can safely say that this game is starting to take on the look of the
far-fetched mental screenshots that inspired it.
28-Jan-2007
8.5
The first parameter of
device->Clear is NOT an MRT index!!!Clear clears all surfaces of the multiple render target.
Did a test to see if interlacing
two anaglyph halves would reduce ghosting (mockups\anaglyph).It didn't, on my CRT monitor.
Realised that the glitches in my
new lighting system were being caused by (LACK OF) floating-point
blending.
Grrr why didn't they support it?!!!We NEED that!!
Noticed that the nVidia 6000 and
7000 cards support floating-point blending (with format restrictions, eg.
maybe only FP16).
My card doesn't seem to support it at all.It can sorta attempt the blend but this causes corruption (flickery
speckles, missing lines, weird colour cutoff) :(
Upped the warning level to 4.
TODO: for reflections, look up
D3DPTEXTURECAPS_PROJECTED.Might be
able to take some maths out of the shader.
Got a series of really
nonsensical crashes trying to run the game.Restarted the PC, rebuilt all and it's been fine (???)
Blending to D3DFMT_A16B16G16R16
(HIGH-PRECISION INTEGER) buffers is not possible, on my card at least.
...but it CAN blend to
D3DFMT_A2B10G10R10.
So maybe the best I can do is use 2TTT for ATI cards and FP16 for nVidia
ones.For the lighting buffers I
mean.
TODO: Should put the FP lighting buffers into an option, for
debugging/cheapening.
Fixed another possible cause for
the reflected lighting corruption: a needless OnResetDevice for the ground
effect, in CLandscape::RestoreDeviceObjects.
Got a couple of nasty hangs, GPU
resets etc. Not overclocked, but ATITool might have been running on each
occasion, not sure.
Got the distorted ground
reflection rendering into the specular buffer (via the final-frame buffer, so
it can MRT-in with diffuse+position+normals).(TODO: skip that detour if floating-point blending is supported).
TODO: Sort the sorting!
27-Jan-2007
9.75
Remember to convert shader
effects in ported code into ESHADEREFFECT entries, otherwise compiler
settings will be wrong!HDRLighting
suffered this, fixed now.
Blackouts hadn't gone away at
all, I had just been hiding them by clamping :(Back to the drawing board then.
I've re-enabled hdr
multisampling and I'm not going to clamp the blackouts for now.Don't want to risk skewing my tone-mapping
results, plus I want to get used to the blackouts to see if I can spot a
pattern.
tone mapping currently quite
well-behaved with *10 specular.*100
it's very glitchy.
Noticed that the HDR code is
always using 'device->ColorFill' rather than Clear.Is it faster?
Added HDR options.
There's now a 'Debug' profile
that only loads in _DEBUG config (overrides 'Default')
New lighting model
(DeferredLighting.fx):
- "gloss" is drawn into normals.a.Gloss allows DeferredLighting to control
specular intensity, specular power, diffuse/ambient intensity.Gloss allows ground.fx to control
reflection strength.The
wetness-darkening disparities between say tarmac and road paint will be done
by editing the textures / patchwork vert colours.Puddles have a gloss of 1.A rough or absorbent material, say wood or
concrete, might have a gloss of 0.5.A
manhole cover might have a gloss of 0.75.
- Reflections will now draw to the specular buffer, where they will be
added to by specular from lightsources.Reflectiveness and specular intensity will therefore become the same
thing, a result of gloss.
- Puddles (rippling etc) will be the job of Ground.fx.The rippling will determine normals, and
the height will influence gloss.Gloss
must not influence water effects, since it's used for dry shiny surfaces
too.
- Raindrop spatters can be animating decals of normal and gloss.
- Env maps will write to the specular buffer, so they don't get treated as
diffuse colour.A pure chrome object
would have black diffuse (for no diffuse lighting), gloss=0 (for no specular
lighting) and would then be coloured purely by its env map in the specular
buffer.
- Rain running down a windscreen would be drips/streams of gloss and
normal.
CModeEnvironment::RenderSubset:
made a possible reflected lighting fix (lighting strength channel might not
have been getting cleared)
Made good progress on the new
lighting system.
26-Jan-2007
3
fxc: I gather the /Vd (disable
validation) flag controls whether or not the compiler enforces the shader
model's instruction count limit.
the define '_TEST' is now
reserved to mean 'in the Test build config'
Read up on vs2005 build
rules.Improved my FX rule to take
custom parameters so that FXC's optimisations are
enabled on release & master.Also passes the config define through rule
parameters.Had to make the
ConfigDefine parameter non-'inheritable', otherwise, bizarrely, the files
didn't inherit the value from the project (??)
Ground.fx went from 139 instructions to 85.
ShadowBlur.fx was failing to
blur the shadow image if FXC optimisations were enabled! (ie. if /Od is
removed).
Fixed this by removing an algorithmic optimisation from its 'for' loop
(revision 218)
Got another instant-reset crash,
when pressing F5 in visual studio, with hdrlighting.fx open.Bootup message said ATI drivers as usual.
...and again, at the beginning
of game startup by the looks of it.
Installed latest ATI drivers.
Installed CS2
Stopped getting blackouts.Had a suspicion they might have been caused
by overclocking, but I can't get them to happen now even with vicious
overclocking.Drivers??(((ShadowBlur.fx????)))
Re-enabled HDR multisampling, working fine.
Dare I say it ... PHEW!
25-Jan-2007
4
blackout hotspot (viewpoint,
look down): x -10151.694 y 757.61230 z 9094.2930
The blackouts were 'caused' by
spikes of CalculateAdaptedLum\fCurrentLum which would normally be somewhere
well below 2.Spike values are very
large, probably infinity or something.
Added a bodge to kill off the spikes.Obviously I'd like to know where they're coming from but I can't be
bothered spending any more time investigating it at the moment.
I'll come back to it later in the project when it's hopefully fixed itself
;)
The 'spectral' glares look best
on white objects, can be slightly mucky otherwise.
Noticed that I've been compiling
my FX files with optimisations and preshaders turned off.<-fixed
Fiddled with the build rules
again, still can't them working properly.
Really have to set up my proper
lighting model now (see 11 dec); the current purely-additive setup is really
naff.And get it behaving nicely with
the tone mapping, blue-shift, (& glares?)
FX build rule(s) currently
broken
24-Jan-2007
5
- disabling projectors now makes the glares appear
- removing the createtexture keeps the glares and some lighting appears, as
do the blackouts
- multiplying-up specular ALWAYS causes blackouts
- *10 caused a blackout, but it seemed more reluctant than 100
- blackouts still happen when flycam is underground
Some run-time warnings/errors to
investigate:
- Direct3D9: (INFO) :No render target available for the given
RTIndex.<-think this is from GetRenderTarget, who cares.
- Direct3D9: (WARN) :Can not render to a render target
that is also used as a texture. A render target was detected as bound, but
couldn't detect if texture was actually used in rendering
(RenderReflectionImage\CProjector::DrawAll)<-was
silly ordering in CModeEnvironment::RenderReflectionImage.
- D3DX: ID3DXEffect::EndPass: EndPass called
while not in a pass
(RenderReflectionImage\RenderSubset\pLandscape->Render)<- invalid texture filter modes (anisotropic) in
bumpy.fx
- Direct3D9: (WARN) :Cannot compute WNear and WFar from the supplied
projection matrix (GenerateDepthImages)
Added ".r" to some
places in HDRLighting where it seemed sensible (was a getting a warning in
max debug mode)
Still can't figure out this one
(bumpy.fx).Doesn't seem to cause a
problem:
D3DX: ID3DXEffect::FindNextValidTechnique: No subsequent valid technique
found.
Gives no warnings at UtilCreateEffectFromFile.<-
invalid texture filter modes (anisotropic) in bumpy.fx
shadow dest, index 1
got an instant-reset crash
(doing a silly temp thing with ORTs)
turned the blackouts into a
1-frame flicker, might be a bit more helpful.
FIX THE BLACKOUTS!
23-Jan-2007
4
Ordered 5 x Red-Blue
glasses.Asked them to give me a
mixture of RB and RG but I don't know if they will...
Update: YAY, they did!Bless 'em.
Started getting this
error:
failed writing c:\Noctambule\Source\HLSL\assembly_Zombie.txt
Project : error PRJ0019: A tool returned an error code from "FXC
compile: c:\Noctambule\Source\HLSL\Zombie.fx"
Tried a different output text file but that made no difference.The compilation is
always succeeding though.
Something about the projector
RTs is causing the missing patchwork base and hdr blackouts (the zbuffer
clear in CProjector::GenerateDepthImages was blatting over other
stuff).
Reducing projector RT size from 1024 to 960 fixes those problems and the
wacky polys (crikey).It's not running
out of vram, definitely not - I can create 12 projectors and it's happy.Tried a bunch of RT formats but nothing
changes it except taking the dimensions down.
It's to do with the HDR's multisampling.Now disabled (temporarily) to get round the
problem.
TODO: fix properly!!blackouts still
present but ground ok
ORTs bigger than the screen
don't have full z-buffer coverage.ORTs don't automatically have private zbuffers either.
Not a fun evening :(
22-Jan-2007
3.33333
Ported the DX HDR sample pretty
successfully :DThe streaky glares are
showing a lot of potential.
Try alt-entering a couple of
times for different glitches.
Trying to track down the
'blackouts':Could they be related to
the stray polys?
- not bad deferred lighting pixels
- not bad ground pixels
- happens when you look around or move
- happens in fly cam too
- setrendertarget z-buffer clear in CProjector::GenerateDepthImages was
blatting over other stuff; device z clear stopped someblackouts.
- doesn't happen in flycam when you face west
- got to be some kind of float
overflow/negatives.Try clamping it to
a sensible (but high) range when copying the frame RT<-no it's not, I tried this
20-Jan-2007
11.75
LogConverter updated for VS2005
and now only writes out the first 30 entries (Devlog.htm was getting far too
big).
A better solution for the devlog is needed in the future.
Didn't mean to, but I spent most
of the day working on anaglyph ghost reduction.Came up with a pretty effective technique
(AnaglyphGhostReduction.fx), but obviously it can't change the laws of
physics so I'm still hindered by the slackness of the filters in my
glasses.My method accounts for a
couple of details I learned today:
- Through the cyan filter, red is much brighter than black, but yellow is
no brighter than green (see 'testPower' param).
- There's a non-linear relationship between how bright a foreign component
is and how much that pixel should be darkened (see 'power' param).
- Adding some brightness gives the deghosting more of a fighting chance.
Currently hardcoded for CYAN-RED.
I'll come back and tie-up the deghosting once I've tried getting hold of
better glasses.
TODO: Order red-green and red-blue glasses (that website kept telling me
"session expired").Try
splicing-up some green-blue ones too.
Deghosting is currently hardcoded to black-out the blue channel.btw this makes it look kinda like a
hologram.
Added a
'stereoscopy.convergence' option to help with ghosting and focussing
problems.
Got all the effects working
properly in fly-cam mode.This
*really* needed doing.
Did some housekeeping in the
camera department.
Got thinking about that
particular type of anisotropic specular that I want.Haven't quite figured it out yet.Checked if there were examples around of
the effect I need, but couldn't see any.
19-Jan-2007
3.5
Had a look at nVidia's
'PracticalPSM' (perspective shadowmap) sample.I'm kinda relieved that it has its fair
share of glitches and limitations.Really I should've found out about this sample before diving into my
own technique.
In particular theirs REALLY doesn't like casting shadows from behind the
viewpoint out in front of the viewpoint.The sample also seems to be taking great advantage of the fact that
the casters are discrete objects far apart - very unlike my town scene.To be fair though, in some situations it
looks very clean, plus it all works with a single scene redraw (single depth
image).Must be tons cheaper than mine
too.
I'm think I'm sticking with my own shadow technique though.
TODO: Right now I'm going to get
the HDR system in place, because I think it'll save a load of faffing with
colours & brightnesses (lighting, reflections, rain) if I can put the
tone mapping in place now.
Airborne/entoptic light
requirements: TODO: Finish planning this
The direction for the game is REALISTIC and
HUMAN-VIEWED.Film/photo camera
effects are a nono, as are game clichés
([bloom/DOF/lens-flare]-for-the-sake-of-it).Above all, keep the scene CRISP not smudgy!That's not to say there's no mist, there
is!
- Use (very) blurred HDR screen image (tone-mapping by-product) to
contribute to the lighting of the rain (/&mist).
- World-space haloes around light sources, revealing rain & mist.Ideally reveals a shaft of rain/mist all
the way from the light to the viewpoint.Use a sort of pointlight refraction shader for this?Nice colour dispersion (redder towards the
outside).This shaft is necessary even
with the blurred screen image influencing rain/mist lighting.
- Razor-sharp eyelash haloes from light sources, preferably also from
reflected light sources & specular highlights too (thresholded fullscreen
star-type filter for the little ones?).Nice colour dispersion around the spikes.Nice counter-rotation as the lightsource
moves across the screen.Blinking
effect?Subtle blinking effect?
Eyelash haloes are non-rain-related and therefore lower priority.They need to be unique to each eye.
- Drips ideally want all the same lighting as raindrops.
- Mist should take the form of waves of vapour in the wind.
- All these effects need to look good in stereoscopic modes.
Rain: TODO: Finish planning
this
In order of rendering:
- Lit rain close to each lightsource, in its halo.Influenced by blurred screen image, and is
also lit by its one lightsource.It's
a world-space cluster around the light + drips in the halo, eg. underside of
street light.
- Rain around viewpoint (foregound rain), lit by/refracting blurred screen
image.
- For each lightsource, apply deferred lighting to the foreground rain in
the screen bounds of the halo.This
can be toggled by a player option.TODO: what about occlusion here?
18-Jan-2007
8.1666
Got anaglyph saturation working
properly, was just a silly mistake.
Started using the C++ 'override'
keyword.It's something I've wanted
since I first started using virtual functions.
Split-off a base class from
WorldParser: BlockFileParser.This is
reused for the options profiles.
ALL options are now read in from
the profiles :D
ORT formats aren't included in the profiles [yet] though.
Got a nice framerate (like, same
as mono) in stereo mode by halving the width/height of the frame RT in
profile 'phil'.
The current bottleneck is very clearly pixel shaders.
17-Jan-2007
3
Visual Studio macros are working
again.No idea what that was all
about.
g_viewInfo.pMainViewpoint is now
generally to be used instead of PLAYER->pCamera->GetPosition().
Removed g_viewInfo.pMainCentralViewpoint.
Shadows now work correctly in
stereoscopic mode.
DXUTBuildValidDeviceSettings:
Now defaulting to double-buffered rather than triple-buffered.This fixes the
'stereoscopy.laggingApproximation' option.
I don't currently know what the advantage of triple-buffering is (?)<- (2.2.7)performance gain apparently; switched it back to
triple
Noticed something weird about
eyes.If I look at a distant object in
stereo mode, then tilt my head (almost on side even), the displacement of the
two perceived crosshairs is still perfectly horizontal across the screen!Seems your eyes are quite happy to
'converge' almost VERTICALLY if they feel the need!Makes no sense that they should know to do
that (???)
Deferred lighting now draws to
the 'frame' ORT rather than the screen itself.
Added an
'stereoscopy.anaglyphSaturation' option.Still isn't quite working somehow - FIX THIS NOW<-done
16-Jan-2007
4
Been away on holiday for a
month.
All Visual Studio macros stopped
working.Can't even run them from the
Macro Explorer.TODO:
Investigate!
Got the anaglyph stereoscopy
working beautifully.Only bug I'm aware of at the moment is that the shadows are monoscopic.TODO: FIX THIS IMMEDIATELY.<-Done
TODO: the
'stereoscopy.laggingApproximation' option doesn't seem to be producing the
correct result at the moment - fix this.<- reducing
number of backbuffers from 2 to 1 fixed this
This week and next I'll continue
with all the bits & bobs I want to get done before I start on the
rain.From the 27th onwards I'll be
practising for the Leeds gig on 9th Feb.Immediately after that I'll get cracking with the rain effects which
are due to be presented in "late May or in June".
13-Dec-2006
1.666
Made a couple of tiny changes to
reflecteddeferredspecular in case they could help with the corrupted
reflected lighting (which they won't).
v7.3 gives different corruption from v7.1/7.2.
TODO: take the code into work and debug it there - I'll be wasting time
trying to fix it remotely by guesswork.
Thinking again about parallax
mapping, having been reminded how easy it can be.No excuse for not putting this in.
TODO: protrusion/relief type shader that includes cheap parallax mapping,
sharpened protrusion edges and per-pixel self-shadowing.
About the edge sharpening though - my placeholder wall textures are very
low-res - maybe I should see what the final res looks like first?
The self-shadowing, er...would have to be done inside the deferred lighting
so it works for multiple lightsources.Need to have a think about that...
12-Dec-2006
2
Bit of lighting & shadow
work.Reflected lighting now uses a
separate effect (ReflectedDeferredSpecular.fx).
Lighting currently has a distance range, but it's just a test.
#version 7.3
Released version 7.3 (*might*
fix the corruption in reflected lighting). Labelled the commit "version
7.3".
11-Dec-2006
2.75
Found that v7.2 actually doesn't
fix the corruption in the reflected lighting on my machine at work :(
I'm hoping the new dedicated reflected lighting FX will fix it (currently
ReflectedDeferredSpecular.fx)
Found and disabled the thing
that was switching my keyboard into French (see language bar 'key settings')
Changed the FX exporter to
include h_Global.fx, which defines _3DSMAX etc.
Put the soft shadows back to
using a simple fade between the source and dest images.They look great that way, but it makes
DeferredSpecular more expensive because of the extra texture sample involved.
Added ambient & diffuse
elements into DeferredSpecular, which look nice.I'm about to try changing the specular
power channel (alpha of position buffer) to be a sort of 'shininess' /
'gloss' channel, partly because I don't have a spare channel with which to
send diffuse data.This also has the
advantage that the interpretation of the gloss is all controlled from
DeferredSpecular rather than in the separate surface materials.Seems very sensible to me.
TODO: DeferredSpecular wants to
add specular to one buffer, while adding diffuse & ambient to
another.Once all the lights have
contributed, the diffuse/ambient layer wants to multiply over the (full
brightness) scene, then the specular layer wants to add over the scene.This will look the business.It will also allow the likes of a torch,
and light-casting weapons etc to work beautifully.
Gord, maybe soft-focussing the ambient-diffuse image would give a nice
real-time radiosity impresion??Sitting underneath the specular and the rain effects, it could be very hard to see how that's been
been done.Could look gorgeous.
P.S. - Yes, I will be having a
1000-hour celebration "field trip".<-done
(june 2007)
10-Dec-2006
5.5
Noctambule's rain (/&
weather) effects have been made the task for my next 'personal review' at
work.
That *should* be due 6 months from now, say 11th June 2007 (added to
Outlook calendar).Would be good to
repeat this 2-birds-1-stone trick next time, since my previous review task
stole 90 hours from the project :(
Options.h: had to name all the
sub-structs to get rid of this link error (??):
LINK : .\Debug/Noctambule.exe not found or not built by the last
incremental link; performing full link
Options.obj : fatal error LNK1179: invalid or corrupt file: duplicate
COMDAT '??0<unnamed-tag>@COptions@@QAE@XZ'
#version 7.2
Released version 7.2 (should
fix the corruption in reflected lighting). Labelled the commit "version
7.2".
With shadowblur.fx in RM, I was
getting a blank preview when I set the PS to SM3.Started working again when I upped the VS
to SM3.So maybe it's a good rule to
keep them matched?TODO: check if the
problem is the same in-game.
Experimented with shadow
softening.The current code blurs from
the source to dest ORT, then reduces the range of the blurred colours in
deferredSpecular.fx according to proximity of the surface with the caster.
Can look nice, but I won't be using this version, because it does too much
damage to the shapes of distant shadows close to their casters (eg. wall
detail shdows are pretty much ruined).
So then, that's the thousand
hour mark passed.
Sounds like a huge amount of time, but it's
"only" 6.3 months worth of office hours.And there's only one of me, and I had no
pre-written engine or toolchain.So
it's not so bad.
Far more worrying is fact that I'm almost exactly half-way through the
project's available time.I started
three years ago! :/
My pace now is far healthier than it was though - I just need to keep it
that way and things'll be fine.
My average pace over the past 3 years was just under an hour a day, which
is terrible.
6-Dec-2006
3.5
Made a good start on the string
table.I think it's a good thing to do
because it forms a sort of design spec for the menus etc. which I should
start adding before too long.
MAX: Added the Map Channel Info
command to the Tools menu.
Got the hang of channel-mapping
my shaders in Max.VertexColour is
working properly now, even displaying vertex alpha in Max (alpha is currently
forced to 1 in game, needs work).
Can't seem to use object colour unfortunately; the 'display vertex colour'
check has no effect when using a shader.
4-Dec-2006
3.83333
Started getting this link error
(see
http://msdn.microsoft.com/library/default.asp?url=/library/en-us/vccore98/html/lnk1179.asp):
Options.obj : fatal error LNK1179: invalid or corrupt file: duplicate
COMDAT '??0<unnamed-tag>@COptions@@QAE@XZ'
Tried fiddling with the 'function level linking' and 'incremental linking'
settings on the project but they didn't change it.
Added a nice 'render target
options' system that will later allow the player to customise ORT properties
from the pause menus.
Disabled warning 4482
("nonstandard extension used: enum
'CRenderTargetOptions::EResolutionMode' used in qualified name")
3-Dec-2006
6.33333
Tried v7.1 on my machine at
work, and all the reflected stuff had crazy corrupted lighting.
Found and fixed an uninitialised texture parameter for deferredSpecular.fx
(if drawing reflections) which might hopefully account for it.
Screenshots of the problem are in the 7.1 build folder along with a
dxdiag.txt.
Started thinking about bug
tracking.
Installed OnTime2006, which required me to install Microsoft SQL Server
Express, but I still couldn't get it working so I uninstalled them
both.
I'll ask at work about Bugzilla...
Found that there's a load of
documentation on the web for Max DX9 shader materials.Why didn't I search for docs
before?!!
Saving these docs in 'docs\3dsmax'
Installed Max 8 and the latest
Panda exporter.Still crashes same as
before.
There is now a version for Max 9, which I'm guessing wouldn't crash.
Here's the info from the panda site:
"Advisory: Dx9 Materials exported under Max8 and Panda DirectX
Exporter may crash due to a bug a 3ds Max 8.Contact Autodesk for details on a fix for this.This bug also affects other exporters using
IGame."
Tried the KiloWatt X exporter
for Max 8, but it makes no attempt to export DX shader materials >:(
TODO: Get Max 9 and try the
panda exporter.
Added whitesquare.tga.
Managed to figure out vertex
colour display for dx9 shader materials in max; see vertexcolour.fx.
Unfortunately Max doesn't set up object colour for non-editable mesh
objects as vert colours (so you can't see vertexcolour.fx on a geosphere node
for example, would have to be solidcolour.fx).It would work on the game side though.
Played about with different ORT
sizes for the effects.NOTE the
patchwork doesn't use MRT, that's why the different targets can have
different dimensions.
Remembered just how ludicrous
patchworkDimpleProcessing.fx is.Addressing this (and keeping its ORT small) will give a huge speedup
:)
Max sometimes just shows random
vert colours with vertexcolour.fx...
...then Panda started playing
up...I'll sort this stuff out
tomorrow.
<- it doesn't seem to like being given a full long file path?Browse to the folder then just give it the
leaf name.
That corruption in the dimples
was nothing to worry about, just the manhole cover dimple object was using
default.fx which I don't include in the game.Fixed now.
28-Nov-2006
0
Spent ages sorting out my email
setup.The compilcations were as
follows:
- 123-reg email forwarding is unavailable because I have to use web-mania's
nameservers for the site to work neatly.
- web-mania email forwarding is to one address each.
- can't access pop3 mail accounts at work because the ports are closed,
etc.
- if sending mail through gmail, there is no way of hiding gmail's presence
in the from field (it does that 'on behalf of' nonsense).Changing the name in the pop account
settings doesn't change that.Changing
the SMTP server to the programmerart one doesn't fix it either.
Email now works as
follows:
- all my selected names ^programmerart.org get forwarded to
allusers^programmerart.org.
- allusers^programmerart.org forwards to
programmerart^googlemail.com.
- programmerart^googlemail.com forwards to the office.
- at home, I pop3 to programmerart^googlemail.com as an incoming-only
account.I've disabled outgoing mail
through gmail and btinternet (should be able to restore this by ticking 'use
same settings as incoming mail server' in the ougoing mail tab).So mails sent from that account by accident
will sit in the outbox and cause an error message till I change them to use
the correct account.
- from home, I send and reply from pop3^programmerart.org using whichever
alias I want (phil^programmerart.org is the default / could use
info~programmerart.org or whatever).pop3^programmerart.org (as phil^programmerart) is my default account
at home.
- at work, I can reply to ^programmerart emails from the gmail site,
whereupon the sender will appear as
"programmerart^googlemail.com; on behalf of Phil Palmer
[phil^programmerart.org]" which isn't too bad.
MSN: when adding a contact, be
sure to un-tick the 'auto-update this contact' option.Otherwise it'll send those request
thingies.
Spent ages faffing with MSN to
try to change my email address.Got it
sorted eventually.
27-Nov-2006
2.83333
Trying to suss this leam.max
crash.
- most recent revision *was* working, so it's not the max file itself
that's wrong
- now versions 127 onwards crash
- ok, version 126 of the HLSL folder makes the latest leam.max work
- it was SolidColour.fx, which had been set to SM3
- REMEMBER: MAX D3D9 MATERIALS CAN ONLY GO UP TO SM2!!
Worked round the above problem
by using the _3DSMAX define to limit the platform to SM2 & exclude the
SM3 code.
FX: Can't use defines in FX
code, eg. can't do
#define PS_TARGET ps_2_0
PixelShader = compile PS_TARGET MyPixelShader();
Started working on shadow
softening
26-Nov-2006
9.25
Deleted ModeSystem.cpp & h
because they weren't being used.Still
in SVN if I ever need them.
NOTE: 'array' is now a C++
keyword.It's a shame, because that
was a useful member name.
Corrected the zombie walk speed
(there was some update in the prep)
Implemented a pause mode (P),
see ModePause.cpp/h.
Missed some serious rainfall
early this morning - woke up at about 6.15 by which time the sky was starting
to brighten which would be no good.The forecast sunrise time was 7.43 - seems the sky brightens long
before sunrise.
Zombie.fx now (properly) shares
code from ProjectorDepth.fx.PS is
selected according to an effect variable, while VS dynamically branches to
call a function in h_Projector.fx, depending on that same variable.
Renamed the most up-to-date
tubby version to tubby.max (renamed old tubby.max to tubby_pre_walk.max)
Had a look at getting the
zombies walking out of step.The
problem is that they're all sharing the same CAnimMesh instance, and that's
where the bone matrices are kept.Would ideally move the bone matrices into CAnimUser, but that'd be
fiddly.
I've decided it's not worth fixing this yet because the current enemies are
only placeholder and will be replaced before too long.
Had to remember to put an
(empty) virtual destructor into CGameMode so that each mode's destructor is
called when ptrs to the base class are deleted.
The player now has a model
(currently the zombie model).It
animates, receives light & shadow, casts shadow, and reflects in the
ground.
Got a crash trying to load
leam.max - TODO: investigate!!!!
Had a nasty bug where there was
a function-scope variable with the same name as a for-loop variable.After the 'for', I was using the variable
for another 'for', and it was picking the wrong one - no warnings.
TODO: Find the setting that will give me warnings for that.
Added VERY SCRAPPY placeholder
zombie-zombie and zombie-player collision.Just stops all that character overlapping that looks terrible.
TODO: check I've got shader
optimisations on in master builds.
#version 7
M tested version 7, working as
expected. Labelled the commit "version 7".
I had somehow lost my _DEBUG and
_MASTER defines from the project settings (along with whatever else was in
there).No idea why that would
be.Could it have been the changeover
to VS2005???The likes of
_UNICODE/UNICODE are inherited.
I've just put the defines back in now, and I'll keep an eye on them.V6 and V7 were missing the _MASTER define,
so they allowed fly-cam to be activated.
Noticed the download size jumped
up a lot between versions 6 & 7; todo: investigate.The only new file was Zombie.fxo.
Brightened the scene back up a
bit (M said it was too dark), and made sure the flycam was unavailable in
master.
#version 7.1
Released version 7.1. Labelled
the commit "version 7.1".
25-Nov-2006
6.4167
Added a SHADER_CODE define
(defined only for the shader compiler).This allows shader code to be placed in headers also used by the game
code, where necessary.
Projector: Got rid of the
PROJECTOR_USE_SHADOW_MATRIX define (now always true).
Finally got the zombies
animating again, with lighting and shadows.
Got rid of that memory leak:
CAnimMesh::pBoneMatrices
Bought some waterproofs, ready
for some heavy-rain photography action.And as a by-product:
I have a cool design for my player character!!
Having the player dressed in waterproofs will be
ace - wonderfully simple and sensible.Can't wait.
They'll be like, "I wonder what I look like, I wonder what I'm
wearing" and they'll go and look at their reflection in a window and say
"Of course, I'm wearing waterproofs, because it's pissing down with
rain.That's genius."
Started getting a memory leak,
don't think it was there on the 20th.Something to do with the change of skinning version for the
zombies?:
{261} normal block at 0x010BD7A0, 656 bytes long. Data: <> CD CD CD CD CD CD CD CD
CD CD CD CD CD CD CD 10
<- fixed (CAnimMesh::pBoneMatrices)
Trying to get the zombies
animating again; dug out the skinning code from SkinnedMesh.fx and planted it
into Zombie.fx.
TODO get this working (genuinely too tired
tonight).
20-Nov-2006
3.75
Fixed the forwarding for my new
domains - it was non-framed forwarding that I wanted.
NOTE: For zombie animation,
search for this line in AnimMesh.cpp:
DEVICE->SetTransform( D3DTS_WORLDMATRIX( i ), &matTemp );
TODO: NEED TO DESIGN THOSE
ENEMIES.NEED TO KNOW WHAT THEY'RE
GOING TO DO AND HOW THEY'RE GOING TO MOVE.
TODO: in bumpy.fx /
solidcolour.fx / zombie.fx, WS positions & normals have to take into
account the OBJECT MATRIX!!!This is
currently only needed for the zombies.
<- done for zombie.fx
Don't press F12 in fullscreen!
Eventually got the zombies
receiving deferred specular lighting, as well as casting and receiving
shadows (this includes self-shadowing, yay!)Looks cool; they look like silver statues at the moment.
It's currently a pretty messy test code-wise - todo: tighten it up a bit.
19-Nov-2006
6.25
Ran into some kind of
too-much-stuff-going-on sorta problem with ground.fx.
Had to keep ground.fx's inverse surface vector calculations in the PS
because of this.
NOTE: Had to change ground.fx STransferData::normal from TEXCOORD3 to
TEXCOORD6 to prevent these errors - makes no sense to me.And there were *definitely* no overlapping
semantics!
error X5639: dcl usage+index: texcoord,3 has already been specified for an
output register
error X4504: overlapping output semantics
error X5629: Invalid register number: 12. Max allowed for o# register is
11.
Moved ground.fx and
deferredSpecular.fx out of RM (renamed to ..._NOEXPORT in rfx backup folder).
SolidColour.fx now writes the
same deferred shading data as bumpy.fx.
Added a PLACEHOLDER modelled
skydome.Remembered that objects only
need the NoctambuleScenic modifier if you want to attach information to them.
TODO?: wheel movement
control.
- roll forwards to move forwards (detects speed etc).
- roll backwards to move backwards (detects speed etc).
- forward & backward movement have sorta momentum so you don't stop
dead between rolls
- nudge in opposite direction to movement to stop suddenly
- hold wheel down and pull left/right to sidestep
Made the ground reflections
distort farther vertically (ground.fx), looks good; todo: improve (currently
hardcoded)
Added deferred specular lighting
(but not shadows!!NO!!) into the
ground reflection image.This makes it
look loads better.
Fixed the occlusion culling in
the ground reflection render, by changing IntersectRayCullPlane so that
effectively every cull plane extends downwards to infinity.
Registered the following domains
for 2 years with 123-reg:
programmerart.org.uk
programmerart.co.uk
programmer-art.co.uk
programmer-art.com
Opted out of WHOIS for the
.co.uk's and the .org.uk; the others don't give me the option.
Set up web & mail forwarding for all the new domains.
Hmm, the framed forwarding
doesn't seem to be working properly [yet] (jumps to the page but doesn't
update the address/title/icon).TODO:
maybe try the other way?
IDEA: enemies could stagger
towards you like zombies until they're quite close, say 10 metres away, then
charge at you - might be quite scary.
18-Nov-2006
6.582
Installed VS2003 and got the FX
exporter working again.Just had to
add the platformSDK and DX paths back into tools\options.
MAX: layer window: press enter
to select
Filtering doesn't work on
floating-point textures??!
TODO: CullPlanes need to hang
below the ground so they work for the reflected viewpoint!!!!
Maybe automate it so they all hang down to the height of the current
reflected viewpoint?
Couldn't get mixed
fixed-point&floating-point MRTs to work, although the DX docs say this
should work.This was for the ground
reflections (colour ORT & position ORT).
RM: VERY IMPORTANT: EFFECT MUST
BE COMPILED IN RM BEFORE EXPORTING IT, OTHERWISE IT'LL NOT INITIALISE
PREVIOUSLY-UNUSED VARIABLES IN THE FX FILE!!!
Got depth diffusion working on
the ground reflections.Looks
nice.Cures some glitches and
introduces some new ones, but I think I'll keep it.Increases the number of texture samples for
the ground :/Uses a new 512*512
floating point RT.
Added a debug RGB multiplier to
bumpy.fx.The landscape and
reflections definitely look nicer (SCARIER) darkened down (before the
deferred specular goes on).
12-Nov-2006
2.5
Had a wcslen on an uninitialised
string, rather than sizeof/sizeof(*).
Had to add bumpy.fxo and
default.fxo
TODO: Find out how to get a
publisher certificate for the downloads.
#version 6
Released version 6. Labelled
the commit "version 6".
TODO: For the next build /
couple of builds, I will be concentrating on AESTHETICS, using the engine
features I've already added (patchwork, reflections, specular lighting &
shadows).Needs to be done, because at
the moment - although these features have great potential - the game looks
mighty crumby.
Amongst other visual improvements, the next build should feature:
- Shadows from all characters (including player)
- Player ground reflections
- Proper (deferred specular) lighting on all characters
- Self-shadowing on all characters
- Landscape improvements (especially ground surfaces)
- Would be good to get the shadow glitches fixed as far as possible
11-Nov-2006
1.75
Fixed the corruption in
fullscreen mode - ORT dimensions weren't getting updated after alt-enter,
duh!
Put the V6 install together
(done for English only so far).The
exe crashes though - seems to be having problems with paths eg. for leam.x
(CMeshFile::Create)
10-Nov-2006
2
Applied for a new savings
account to use for the charity proceeds from the project.No point letting JustGiving scoop 5% off
any donations, cheeky buggers!
TODO: add this to PayPal; put a donate button (plus info) on the site.
Stream mappings are not exported
in the FX file and only apply within RenderMonkey!
MAX: trying to find the semantic
if any on which vertex colours are passed.Tried COLOR0, COLOR1, COLOR2, TEXCOORD1, TEXCOORD2, NORMAL1,
POSITION1.No luck.
TODO: TRY THE SAME IN MAX8; see VertexColor.fx, which suggests that vert
colours come through on TEXCOORD0(?)
Got a load of problems with
SolidColour.fx when I renamed __Diffuse to something else.The name needs to contain '__Diffuse' in
order for the semantic to get added by the exporter.Weird thing is it still didn't work (didn't
write out the semantic or any XML type stuff) after changing it back
from__Colour to __Diffuse.Had to revert the RFX and add __Diffuse
again.(???)
<- just need to remember to compile before
exporting; see 18 Nov
TODO: Get VS2003 so I can edit
the FX exporter again!<-DONE
Alternatively, could maybe try a recent version of RenderMonkey, see if it
has a 2005 exporter and if so how much work it would need.
9-Nov-2006
4
TODO: Add the ability to draw
meshes with whatever effect (eg. ReflectedLandscape.fx).
Dunno if it might be a matter of having them as CEffect / CEffectInstance
instances so they can be fed textures and that…
Currently using Bumpy.fx for the reflections, temporarily.
TODO: Get vertex colours working
on Default.fx (in Max and in game).It's being used on those light marker spheres.
<-Vert colours really don't seem to be passed
to the shaders in Max 6; try Max 8 at some point.
8-Nov-2006
4
Been busy with an effects
post-mortem for work, and with band stuff.Also got severely reeled-in to MySpace, horrible thing, I've quit now.
Rendered my TV unusable,
informed TV licensing and cancelled my licence payments.
Finally got the hang of
housework scheduling: half an hour every day - no more & no less.
Got a new passport.
Stopped drinking, until at least
2nd Oct 2007.
Also gone veggie and started
exercising every day.That's not
really relevant but is part of the whole self-discipline drive which I'm on
for the benefit of the project.
ALT-ENTER: looks like 72 or 211
refs on the device when I reach the reset (Debug D3D)
CAnimMesh.cpp: Moved a block of
code from RestoreDeviceObjects back into InitDeviceObjects (alt-enter was
revealing this as a leak on game shutdown)
ALT-ENTER: added a
LANDSCAPE_ENABLED define; setting this to 0 makes alt-enter a-ok.So whatever the problem is, it's inside
landscape.
ALT-ENTER: Added
PATCHWORK_ENABLED define.Shows that
patchwork has nothing to do with alt-enter.
ALT-ENTER: Added
LANDSCAPE_MESHFILE_ENABLED define.Shows that the meshfile has nothing to do with alt-enter.
ALT-ENTER: Added
LANDSCAPE_MESHOBJECT_ENABLED define.Shows that the problem is only present when mesh objects are created.
ALT-ENTER: Added
RENDERABLES_EFFECTINSTANCE_ENABLED define.Shows that the CEffectInstance has nothing to do with alt-enter.
ALT-ENTER: Added
RENDERABLES_MATERIAL_ENABLED define.Shows that the problem is only present when CMaterial objects are
created.
ALT-ENTER: Added
RENDERABLES_EFFECT_ENABLED define.Shows that the problem is only present when CEffect objects are
created.
ALT-ENTER: Added
RENDERABLES_D3DXEFFECT_ENABLED define.Shows that the problem is only present when ID3DXEffect objects are
created.
OnLostDevice IS getting called
for ID3DXEffect instances.Are there
maybe also some instances that aren't getting the call?
D3DXFX_NOT_CLONEABLE (as in
statemanager sample) when creating the effects didn't fix it.
Something to do with
'getExisting'?YES
IT WAS.
ALT-ENTER: Here's what it was:
In CEffect::Create, 'getExisting' was being passed raw effect names, not
adjusted effect names.So probably the
effects with a 'MAX_' dummy version would've been the ones tripping things
up?
All working lovely now, apart from some weird corruption at the bottom of
the screen in fullscreen mode (where the task bar was).
TODO: Fix fullscreen corruption
at base of screen.<- ORT dimensions weren't getting
updated after alt-enter, duh!
#revision 124
8-Aug-2006
2
Managed to untangle my source
control.
Various SVN faffing.
Made a DVD backup of the
Noctambule folder.Seems I either
forgot to do this last time or I did it but nothing wrote to the disk.
See Nero was somehow set to use its 'virtual recorder' rather than my
physical DVD recorder (?)
#backup 1
6-Aug-2006
2.5
Release-mode startup crash:
Narrowed this down to the 'Basic Runtime Checks' setting being 'Default'
rather than
'Both (/RTC1, equiv. to /RTCsu)'
In fact it's some problem finding leam.x in
CMeshFile::Create.Search for
D3DXFERR_FILENOTFOUND.
TEMP BODGE: I've hardcoded-in the leam.x path
in CMeshFile::Create, which gets round the problem.TODO - FIX THIS!!!Is it just a problem with the initial
directory?
Added a 'Test' config that can
be used for any old config testing.
#revision 115
Tried bailing out of
OnFrameRender altogether but the state block problem persists.
5-Aug-2006
5.16
ALT-ENTER:
- Should only delete device objects (DXUTCleanup3DEnvironment) if something
goes WRONG (which it is currently).
- Should just reset the device (DXUTReset3DEnvironment).
- Can (and should) reset the device with a non-zero reference count.
- My device reset fails with D3DERR_INVALIDCALL, need to understand
why.Shouldn't fail.
Debug D3D gives me this output
on the Reset:
"Direct3D9: (ERROR) :All user created stateblocks must be freed before
Reset can succeed. Reset Fails."
TODO: NEED TO TEST ALL BUILDS ON
DEBUG D3D BEFORE RELEASING THEM!!!
V5 crashes on Debug D3D.
Switching off D3D 'break on…', I
get a bit more output but no more info:
"Direct3D9: (ERROR) :All user created stateblocks must be freed before
Reset can succeed. Reset Fails.
Direct3D9: (ERROR) :Reset failed and Reset/TestCooperativeLevel/Release are
the only legal APIs to be called subsequently
D3D9 Helper: IDirect3DDevice9::Reset failed: D3DERR_INVALIDCALL"
NOTE: IDirect3DDevice9's
reference count member, in DEBUG D3D seems to be a signed int 4524 bytes in.
In debug d3d, device refcount
always reads 211 when I get to the alt-enter device reset.Release 1464, hmm.
Tried disabling font creation;
didn't affect the alt-enter stuff.Put
the font creation back in the way it was.
NOTE: Release build currently
crashes on startup.CMeshFrame::FindFrame doesn't find the frames the game needs.No idea why.TODO: Fix this!!
Currently the best way I've got
of doing a printf Is by using
CRT_DBG_REPORT(_CRT_WARN, NULL, 0, NULL, "hello %s",
"world")
This only works ifdef _DEBUG (annoying, because release buids are where I
need it most).
printf doesn't work at all.
Set up a macro key for the above, ctrl-P.
2-Aug-2006
3.5
Trying to get alt-enter
working.At the moment I'm not sure
how V5 managed alt-enter.By comparing
the current source with the V5 source (eg revision 26 of the Source folder) I
should soon be able to work it out…
Remembered that there's currenty
sorta two landscape mesh systems sitting side-by-side, need to fix
that.
The newer one from StateManager only stores like one CMeshObject,
representing the top-level leam.x thing…
The rest is done by CMesh.
30-Jul-2006
7.2
Trying to track down these
unreleased device objects.
- always three of them.
- nothing to do with Projector, although I thought the problem started
about the same time as Projector went in.
- I found exactly three manual device->addRef calls, but they all
release as they should - was just coincidence.
<- Gotcha.CModeEnvironment::RestoreDeviceObjects was doing pre-CRenderTarget
stuff to setup 3 render targets.
NOTE: IDirect3DDevice9's
reference count member, in RELEASE D3D seems to be an signed int 48 bytes in
- handy for data-breakpoint action, to see what changes the refcount.
*((int*)(void*)(int(g_game.pd3dDevice)+48)) // release d3d
*((int*)(void*)(int(g_game.pd3dDevice)+4524)) // debug d3d
#revision 110
TODO: Disable the F3, F8 & Pause DXUT hotkwys.
TODO: From now on, after adding
any feature, must test that alt-enter still works and that the game shuts
down properly!!!
Fixed a few wide-string-length
bugs; they wouldn't have been causing a problem though.
NOTE: DEVICE->GetRenderTarget
won't output NULL for an unused stage; it'll just not touch the pointer you
pass in.Will return
D3DERR_NOTFOUND.
TODO: Manually keep a record of which stages are in use.
On alt-enter, with maximum
validation etc:
"Direct3D9: (ERROR) :All user created stateblocks must be freed before
Reset can succeed. Reset Fails.
Windows has triggered a breakpoint in Noctambule.exe."
This would seem to tie in with the D3DERR_INVALIDCALL return code that I
get from the Reset.
#revision 111
#revision 112
RMB is a debug key to
Invalidate,Delete,Restore. TODO: Get this process working (landscape is
causing problems).
TODO: Get alt-enter working.
#revision 113
Website: Deleted the tracker
because it gave a nasty offputting 'active content' warning whenever someone
visited the page.I don't even know if
hits were being recorded if the visitor blocked the script.<-only
happened when opening the local harddrive copy of the page
29-Jul-2006
6
Class 1 lights: Maybe only ever two of these at a time.Use generous ORT res and res bias.Can cast live shadows.Shadows may undergo softening process.Draw over whole screen most likely.
Class 2 lights: Say five or six of these, farther away.Small ORTs, simple projection (90° cube
style).Depth images (shadows) are
frozen.Deferred lighting & shadow
done using a screen-space revealing sprite.
Class 3 lights: Can have loads of these, way off in the distance.No shadows.Deferred lighting done using a screen-space revealing sprite.Consider limiting lighting influence to a
WS range if lack of shadowing becomes offensive.
I'm going to put together a
build with a couple of class 1 lights just to show some progress; once that's
done I'll work on the full-blown light management system.
NOTE: Max does not like it if
you set Windows' number style to French.The spinners can't make sense of the commas.
In my FX build rule, I tried
changing the output paths to use backslashes instead of forward slashes - MB
suggested this because he'd had some similar problem.This didn't fix mine though.
Added a 'Draw to shadow' flag on
the Scenic modifier (defaults true, handy to set false for light marker
spheres - which are temporary of course).
Verified with Fraps that
backface-culling the building walls gives a significant speedup (eg. 43fps
-> 48)
Found that I was getting a load
of WM_TIMER messages, which were causing error output when passed to
DispatchMessage (DXUT.cpp).I can't
see any Windows timers being installed though (a function called SetTimer
would be used for this).
I'm now forcing it to ignore any WM_TIMER messages, which gets rid of the
output (and presumably any related file access).
The error was:
'Noctambule.exe': Loaded
'C:\WINDOWS\system32\psapi.dll', No symbols loaded.
'Noctambule.exe': Unloaded 'C:\WINDOWS\system32\psapi.dll'
Looks like the D3D sample apps suffer from this
too, hmm.
17-Jul-2006
Got rid of the SayBox because it
stopped working.
16-Jul-2006
3
The 'giant gap' in the shadows
is caused by a fault in the projection matrix.I've got no idea how (the calcs all look
fine), but it's like the fovs or the aspect ratio goes wrong.TODO: FIX THIS!!!!
As a temporary bodge, I've artificially widened the
aspect ratio.REALLY need to fix this.
Brought a bit of res bias back
in, because it does help reduce depth discrepancies as long as you don't use
too much.
Decided to ignore the shadow
glitches for now so I can push on and get v6 out to show some shadows.I'm going to lose too much morale
otherwise.
TODO: get visibility testing
working for depth images
9-Jul-2006
2.5
Don't know why projecting the
depth images onto the shadow matrix causes red gaps on the depth image at the
frustum's far end.It's the shadow
matrix transformation itself that causes it, not the res-bias adjustment.
Tried disabling D3DRS_CLIPPING; that didn't help (saw no difference).
Tried various near & far clip settings.
Setting the near clip to 0.f fixed it :D
Unfortunately, seems res-bias
was responsible for a fair chunk of the depth discrepancy problems I was
having (it takes res away from stuff farther from the viewer; less res = more
discrepancy).Maybe larger depth orts
are the way to go?
Currently shadow matrix and res bias are disabled.Might be able to put them back if the
normal-based depth shift described above is successful.
TODO: got to figure out what
that giant gap in the shadows is when looking towards a light.
8-Jul-2006
4.25
Using 1024*1024 shadow depth
ORTs caused red borders at the bottom of the images (1024 must be taller than
the window surface).
Made frustum faces at very
shallow angles to the projector not draw (gets rid of some hairy overlapping
nonsense).
In the absence of a 16-bit
one-channel integer ORT format, I'm currently using R32F for projector depth
rather than G16R16.
Important: I'm still doing the depth division though, to keep numbers
closer to zero and therefore more accurate.
Seems you get no filtering using floating-point textures??!!!I'm using G16R16 for now.Filtering really helps prevent depth
discrepancy.
NOTE: Higher-res depth images ==
significantly less depth discrepancy.
The remaining depth
discrepancies seem pretty honest - the main one is that surfaces at very
shallow angles to the light don't have enough pixels representing one of
their axes, so the gradation of reference depth is lousy.
TODO: In ProjectorDepth, bias verts' depths to be
farther back according to the dot of their normals with the light's view
direction.
7-Jul-2006
2.75
Turns out alpha blending is not
supported on floating-point render targets on PC.Got some docs from F about workarounds etc.
So I tried some integer
formats.16 bits would be plenty
resolution. D3DFMT_L16, D3DFMT_A8L8 and
D3DFMT_A8 gave me a scary "Could not find any compatible Direct3D devices"
error & crash. D3DFMT_G16R16 didn't support
blending. D3DFMT_R8G8B8 turned the game into a
black void.
Remembered that pixel shaders
can output arbitrary depth values - now using this and zbuffering, rather
than trying to use the MIN blend.
Got loads of depth-discrepancy speckles trying to use a floating-point
format though.
Now working pretty well with D3DFMT_G16R16.
TODO?: use a depth-only format?
6-Jul-2006
1.666
Couldn't get projector depth
images to use blending when their ORT format was floating point.
4-Jul-2006
3.5
Got the dynamic res bias
working.Had to limit it at about 0.8
because when it gets extreme it gets glitchy, which is fair enough.
Plenty of depth discrepancies
caused by the shadow matrices etc.
TODO: record the depths AFTER those transformations (in both ProjectorDepth
and ProjectorShadow)?
2-Jul-2006
8.5
Realised that res bias will not
help the PZ face much at all, especially when looking away from the
projector.Should still be applied
though, I think.Might well want to
use higher-res targets for the PZ faces.
Decided to avoid
flow-control/branching problems altogether, by drawing a separate pass for
each frustum face.
Will probably end up drawing these shadows to a smaller-than-screen size
ORT which will then be blurred and look-up by the deferred specular draw.
Note: Lights-out coding is good
- feels more determined and reclusive.
FINALLY res bias is starting to
work :DNeed to get the correct amount
applied based on size-on-screen of the two face ends (currently, PZ face is
kinda buggered by it).Looking really
nice though.Very sharp (but short)
sawtooth edges are produced; these might actually be beneficial when I come
to blur the shadow image.
I'm so glad this res-bias thing is finally starting to bear some kind of
fruit.PHEW!
1-Jul-2006
4.5
HLSL: dividing zero by a value
can cause a bad result???
This weird DeferredSpecular
cutting-out / corruption - looks like it could be some kind of parallelism
problem?It's almost as if it's using
values before it's finished calculating them or something.
When I try to combine
referenceLightDepth values with debug colours from the array (looking up
using index from GetIntersectedCubeFace), I get problems.Either on its own is fine.Using a constant (hardcoded) colour index
is fine.
If I assign the appropriate
debug colour to a temporary colour variable next to each tex2D call, I can
combine the referenceLightDepth and debug colours, but only if I don't divide
the referenceLightDepth or the debug colour (except by 2).
I get slightly different results
depending on whether I use /Gfp or /Gfa in the FXC compiler arguments.Both look about as wrong as each other
though (both cut out, just at different times).
Looked up 'flow control' in DX
help and found there are some limitations to be aware of.Didn't totally understand it, but it did
sound like it could be relevant to the problems I've been getting.
Fixed the frustum plane normals
(had been using slightly the wrong trigonometry).
Couldn't work out why the
projector fovs seem slightly off.The
geometry that calculates them seems correct.Is it just float imprecision??
TODO: RES BIAS!!!!!!
29-Jun-2006
3
Installed Office 2003 (from L).
Right, turns out that
DeferredSpecular cutting-out weirdness was some problem with the NZ face
matrix, even though it looks healthy in the watch window.I reckon I don't actually need that NZ
face, it's so tiny.I'm going to carry
on without it and see how it goes.It
did seem like a waste of frame time and memory to be rendering a scene to
such a tiny area.
TODO: try to make sense of those
>1 U coords, on the NX face for example. <-Was treating the frustum box as 1*1*1
rather than 2*2*2.
28-Jun-2006
2.5
HLSL: DON'T TRUST 'clip' WHEN
THERE'S CRAZY STUFF HAPPENING (?)
Eventually managed to stop the
shadows cutting out.
- face matrices' translations
were in slightly the wrong space.
- a UV needed fmodding with 1
otherwise the whole screenquad would cut out.
- don't trust 'clip' when weird
stuff is going on.
TODO: get 'lightDepth'
calculated; get other faces mapping properly; get shadows working properly
(depth comparison) again.After that,
do the shadow-matrixy stuff res-biasy stuff...
26-Jun-2006
3
Signed up for BT broadband;
should arrive at work on the 11th of July.They now reckon I can get speeds up to 6.5 gig?
Tried to do some shader
debugging again, couldn't get it to run successfully.
Tried to make sense of this
cutting-out lark that DeferredSpecular is now doing.It is properly bizarre - not a code bug,
something else is going on.HLSL
optimisations and preshaders are currently disabled, but this makes no
difference.
25-Jun-2006
9.85
IMPORTANT: Frustum-space
direction vectors are tricky buggers.See HLSL function 'GetFrustumSpaceUnitDirVec'.
Set up syntax highlighting for
HLSL.Set the 'editing experience' for
FX files to be C++.Put usertype.dat
shortcut on quick launch bar.
IMPORTANT: I'm editing
DeferredSpecular.fx directly from now on.RM was holding me back (wouldn't let me use arrays).
Finally got shadows landing in
the right place.The colour is cutting
out though, maybe something to do with W's.Current debug shows zero W's in red and
non-zero W's in green (????).
TODO: Get latest drivers for
DVDRW, backup Nocatmbule folder!!!!!!!!!!
24-Jun-2006
9.666
NOTE: RM: SEEMS MULTI-LINE ARRAY
DECLARATIONS CRASH THE EXPORTER.
NOTE: RM: SEEMS ANY ARRAY
DECLARATIONS CAN CRASH THE EXPORTER.
**** IMPORTANT **** D3DX: WHEN
MULTIPLYING A VECTOR BY A FUNKY MATRIX (eg. perspective, probably shadow
too), NEED TO USE D3DXVec3TransformNormal NOT D3DXVec3TransformCoord!!!!!
Eventually got the frustum-space
thing worked out.It's starting to
look sensible now.
23-Jun-2006
2.333
Made a fix to projector view
mats: target pos was between top&bottom points rather than left&right
points.View mats are looking right
now :)
PROBLEM: ProjectorDepth: might
need to do all the transformations (shadow matrix, perspective trickery) in
the PS, because doing it in the VS will change the Gouraud (different
perspective) and hence the depth information.
TODO: trace ray from each pixel
in the direction of (light->pixel); find which frustum face it hits so we
know which texture to look up.Each
face will have a view matrix associated with it; multiply pixel's WS position
by that view matrix to find where on the depth image the pixel maps to.Bonne chance !
21-Jun-2006
4.57
Made a good start on the shadow
res-bias thing - it's looking promising.
NOTE: ProjectorDepth.fx is
currently bodged to help debugging the res-bias.
NOTE: Hid pointlight01 marker
sphere because it was getting in the way.
NOTE: CMesh visibility testing
is currently bodged to always return true.
TODO: Flatten depth image using
shadow matrix, then adjust plane so perspective expands narrow end of frustum
face as required.
20-Jun-2006
1.5
Got debug lines working again
(screen quads had been writing to zbuffer; UtilScreenQuad now sets zwrite
false).
Checked that the frustum plane
data are all correct.
TODO: When I come to fitting the
frustum faces into the ORT views, draw each frustum face in WS to check it's
fitting cleanly.
TODO: Res bias technique is
currently X only; need to get Z in too, somehow.Maybe using perspective to expand the near
edge is indeed the way to do it (bring near edge closer to light till it
fits)?
18-Jun-2006
5.5
Got the sky colour sensible
again after adding the deferred specular.NOTE: device->Clear clears ALL MRTs.This was happening inside
CPatchwork::GenerateTextures.
Started working on the shadow
res bias stuff.NOTE: debug lines
currently aren't working.
TODO: GET DEBUG LINES WORKING;
CHECK THAT THE CPROJECTOR VIEW FRUSTUM STUFF IS OK.
11-Jun-2006
10.5
RM: After changing an effect,
you need to either F6 (faster - effect must be current though), or save,
before exporting.No change otherwise.
IMPORTANT: FLOAT VALUES LOSE
PRECISION THE FARTHER THEY GET FROM ZERO!!!
Therefore it's best to subtract
the VP from WS positions before storing them in an RT.
Spent ages trying to work out
why my depth image colours were off; turns out I'd just set up some wacky
blend mode in the ProjectorDepth effect >:| Or rather, that the blend
didn't account for the fact that the target wouldn't be cleared to a sensible
colour.
Had to disable dithering in the
ProjectorDepth effect to prevent artefacts if using D3DFMT_R16F targets for
it.With D3DFMT_A16B16G16R16F targets
it wasfine (?)
Can't go full-screen currently.
Got light colours passed through
to deferred specular.
TODO: GET FLY CAM WORKING AGAIN;
see end of CProjector::GenerateDepthImages.
10-Jun-2006
7.15
Got the deferred specular
working, YUSS!Had to normalize the WS
normals per-pixel.
Turns out you CAN store any
values in floating-point render targets.
IMPORTANT: RM does not like
dealing with any compile errors while multiple render targets are in effect
(resets the machine, something about ati3daug.dll).
Tried to get the FX exporter
project to build in vs2005, couldn't do it.Bloody unicode bullshit again.I suppose I'll have another go at some later date.
RM exports the
'SEPARATEALPHABLENDENABLE' renderstate typo'd as
'SEPERATEALPHABLENDENABLE'.Really
handy that, cheers.
Started to get some patterns
that almost resembled the shadows.For
some reason the WS position image is seriously lacking in rgb resolution (?)
TODO: Stop the ground reflection
drawing into the projector depth image.
TODO: Address the ws position
blockiness somehow.
TODO: Get those shadows
a-workin'.
5-Jun-2006
1.24
Had to append "_Tex"
onto the name of ORT effect parameters when retrieving handles, because the
FX exporter appends that.
Remembered that
UtilDrawScreenQuad sets up an ortho view matrix.If it's used in conjunction with a shader
that doesn't use the view matrix...well that was causing my deferred specular
to come out crazy.Now added a
'setMatrices' parameter.
TODO: Fix this current deferred
specular problem; compare dummyScenery position colours with position colours
in game, they're different.Looks like
RGB can be stored outside the 0..1 range?
2-Jun-2006
2
Had trouble using UtilScreenQuad
with a texture on it; ended up remaking it basing it on
CModeDebug::ShowTexture; now works fine.
Made some progress with deferred
specular.
30-May-2006
2.75
Installed Tortoise SVN, nice
Windows integration for SVN.
Had some difficulty removing
folders from source control.Ended up
having to use the command line (svn delete) and then the right-click menu
'commit' option.
29-May-2006
4.6
Added a linked list class
(CLinkedList, CLinkedListEntry).
Added a render target wrapper
class (CRendertarget).
Made CNoctmabuleApp into a
gamemode (the very bottom-level gamemode).
CNoctmabuleApp (g_game) now owns
all the render targets - much neater.
Decided I want to render the
scene to an offscreen 16F render target, for HDR.Need to render to such an ORT anyway so I
can use MRT for the deferred specular images (ws position, ws normals).
28-May-2006
9.75
Decided that shadow-casting
lights will apply their light to the scene in a deferred specular pass over
fullscreen images of each pixel's WS position and WS normal.This includes the ground
PROS:
Means I don't have to worry
about assigning lights to materials.Which light reflects off what is a screen-space matter.
Provides vertical streaks for
the lights in the ground reflection.Can even fortify these by biasing the specular reflections in the Z
axis ("to what extent is the vector from the pixel to the light VERTICAL
from the player's point of view?")
Most importantly, lets me use
the shadow image to block out only the light casting the shadow (not darken
the ground reflection indiscriminantly like other games would).
CONS:
When rendering the reflection
image, will have to duplicate the whole deferred specular process if
secondary reflections are desired.
Hopefully shadows should provide
suitable occlusion for the light reflections; if not it might look a bit odd
that light can be seen reflecting 'through' objects.
Will want some code to decide
the screen area in which each light reflects.
Realised that the notion of
'shadow range' doesn't make sense for specular lights.Luckily my res-bias method should make
limitless shadow range possible.
Had some trouble with this error
in RM:
RENDERING ERROR(s): Invalid dimension or type in
shader constant parameter named 'viewPosition' in pixel shader 'Pixel Shader'
in pass 'DummyScenery'
It made no sense.In the end I just added a fresh float4 THEN
added the vViewPosition semantic to it, and that worked fine!Must simply be a bug in RM.
RM: CAUTION: Remember that the
'modelBoundingRadius' semantic uses the radius of the CURRENT PASS model
(might be like a 1x1 square or something).
NOTE: Seems even on
floating-point render targets, you can only write colour values between 0 and
1 :(
Removed the .suo and .ncb files
from subversion source control - they were making a nuisance of themselves
Maybe the best thing to do is
use dummy versions of the shaders for Max, that take all the same parameters
as the real ones.
That's working now - just put
"MAX_" at the start of the name of the Max version of the effect.
RM: got an export crash in an
effect with a disabled first pass and an unused texture.Deleted them both and one of them fixed it.
BumpyWall.fx is still referenced
somewhere in the world but I don't know where (it's in a non-exporting
material).
To get round problems, I've made
BumpyWall.fx a relic (write-protected).It should not be altered and is superseded by 'Bumpy.fx'.
27-May-2006
8.75
Chose D3DFMT_R16F for the shadow
depth images.Might consider using
DF24 DST later for cards that can use them; see ATI 'ShadowMapFetch4' sample.
Haven't figured out how to
switch to my onboard 6100.Managed to
disable the Radeon though - wasn't a very smooth process.
Re-added all my .FX files to the
project so they use the FX build rule.For some reason it now only compiles one effect per build (???)
26-May-2006
1.666
Worked out how to distribute the
res of the shadow images.It's going
to be awesome - people will never work out how I did it.
Started on the CProjector class
which I'll use for all the shadow-casting lightsources.The job of CProjector will be to calculate
the planes for the cube map faces, which ones need to be drawn, and the
matrix required to fit each one to a square image.Shaders will take over from there.
TODO: Ensure I'm clear about how
to map the shadow images onto the environment.
TODO: Continue with CProjector
(start with downwards face of Pointlight1).
24-May-2006
4.25
- SHADOW/LIGHTING DESIGN is top
priority.It's something that's best
planned out now rather than later.
Decided to use projected depth
textures for the streetlights (see ATI ShadowMapFetch4 sample).Sod sub-1600 cards, for the mean time at
least.The technique can be made to
work on any old card later.
The main challenge is going to
be hiding the aliasing.The ATI sample
looks nice but it's using a 1024*1024 ORT, and focusing just on that one
object.Some stuff to try:
Can the depth image be
blurred?I don't see why not.
Lightsources could take
advantage of the fact that their whole range isn't always in view.
Since I'll be updating the depth
images per frame, could the lightsources render depth images of geometry
transformed by OUR view then by the LIGHT's view, to bias the res towards
receiver geometry that's larger on screen?The extra cost would be in the VERTEX shaders I'm guessing, which
would be nice.
Somehow allocate more res
according to each lightsource's range-size on screen.Make sure lights at the far end of a street
aren't actually rendering shadows, just doing simple light casting.As they grow big enough on screen to cast
shadows, fade smoothly from their plain light spot to the spot with shadows.
Will probably need five-faced
(cube map) depth images for the general case.
Maybe casting lightsources
farther away can be non-dynamic, and just hold onto static low-res depth
imagesfor as long as they're on
screen.When they grow big enough to
get dynamic, an off-screen cross fade from that to the full-res dynamic depth
image can be performed.
REMEMBER: each pixel in our
screen space can work out its texcoord in light space and its depth in light
space (while it's being drawn to our screen space).
Could maybe do some screen-space
sweetening of the final shadow image?
**** TODO **** GET CRACKING ON
THIS SHADOW/LIGHTING WORK IMMEDIATELY.IT'S GONNA BE A BIGGY.
TODO: TAKE SHADOW/LIGHTING
REFERENCE PHOTOS ON A RAINY NIGHT (TOMORROW NIGHT?)
Examined both the POM
demos.Decided this is worth putting
in at some point.Actually seems very
affordable when the sample limit is low, which it can be if the depth doesn't
need to be extreme.For the walls,
probably best to perform the POM as deferred rendering, making it easier to
use depth decals?This will reduce
image quality a bit, but that's probably not a big concern.POM should really jump to life in
stereoscopic mode - can't wait for that!
Decided I *do* want some
fogginess in the town.It just looks
cool.The enemies will look a bit
scarier if they're sorta shrouded in fog in the distance.
Added sewage cover to the DX POM
sample: file://C:\Documents and Settings\user\My Documents\Visual Studio
Projects\ParallaxOcclusionMapping\ParallaxOcclusionMapping_2005.sln
23-May-2006
2.85
Got Subversion working; the
Noctambule folder is now under Subversion source control (only the stuff that
needs to be though).The repository is
at C:\Repository.Useful commands are
'svn commit' and 'svn diff'.
Got the level working again - it
was just the reshuffle of losing the 'VSS' folder that had broken the DX9
materials.Made a 'Default.fx' to help
with that kind of problem in future (that's the default effect name used by
DX9 materials).
VS2005 'build rules' are
good.Used one of these for my FX
files; currently only handles debug config though.Keeping the rule files in
Noctambule\VisualStudio\BuildRules.
18-May-2006
2
Installed Visual Studio 2005
(from C).Took a copy of the install
DVD (see 'Installs' folder).
Did away with the 'VSS' folder
because it should never have needed to be there (was to do with a SourceSafe
quirk).
Got the game running and
debuggable in VS2005 :)
TODO: Set up Subversion ASAP.
TODO: Debug the current Leam.x;
get it working (probably just a minor materials problem).
Eclipse said my X1900XTX arrived
into stock today.Should be
here...tomorrow maybe?
So, the Noctambule project is up
and running, at home, finally - w00t! :D
16-May-2006
2.5
Got Max set up.
Got RenderMonkey set up.
Did a bit of landscape
modelling.
7-May-2006
5.5
RM: Stream mapping:
-seems to correspond to vertex declarations.
-affects RM previews only.
-does ordering matter?
-X mesh colour is in 'COLOR' format, not 'FLOAT4'.
Keep getting this from
RenderMonkey when I try to introduce colour streams:
DirectX Preview Window: Error setting up vertex data
for model 'ScreenAlignedQuad': channel 'Color_0' data is missing from the
model.
With X meshes, got to set the colour stream
format to 'COLOR' not 'FLOAT4'.
Tried using an X mesh in RM, haven't managed to get it to work so far, but it would be a
good idea to use X meshes for previewing - rule out any file format
differences.
Need to use the 'no frames' export option for
RM preview meshes.
NOTE: The easiest way of
checking what does & doesn't come through in a mesh file is to use Deep
Exploration.
But bear in mind that it can be a bit lazy about updating the preview when
the file has changed.
NOTE: X
exporter swaps Max's Z&Y axes.
FVF CODES ARE A PRE-D3D9 RELIC
AND THEY DON'T WORK - USE VERTEX DECLARATIONS INSTEAD!
Changed the mouse control so
that push-forward is look-up.This
seems to be more popular.I'll
obviously add an option later.
Installed April 06 DirectX SDK
Released build 5, the first
build to feature puddles.
I think I've worked out how to
get mesh vertex colours from max into RM and the game - yay!
I'm quite inspired by these
zombies:
http://ishi.blog2.fc2.com/blog-entry-180.html
Would they work in 3D I wonder?
Also the ??? tool is interesting reference for the dreamy effects.
6-May-2006
7.5
RenderMonkey: F5 to compile
current SHADER; F6 to compile current EFFECT.
Shaders: Watch out for those
bloody D3DRS_WRAPn renderstates, they're confusing and hazardous.
If in doubt I think it's best to keep them at zero.
Forgot about the perspective
problems involved in mapping a ground ORT onto WS ground geometry; wasted a
fair bit of time in PatchworkDimpleProcessing because of this :(
Turns out my dimple surfaces
were in the correct 'plane space'; they just had inverted binormals!
Always check the obvious!
Binormals point DOWN, negative local Y.
Fixed the 'sliding' puddles -
the dimple lookup in Ground was happening too late, after the uv vector had
been adjusted for a different type of lookup.
Started getting some weird
framerate jittering (all configs).There's a slight reqular jerk about once per second.
<- restarting the PC fixed this (?)
Tried to minimise the number of
pixels drawn in PatchworkDimpleProcessing - very important to do this kind of
thing, big optimisation.
TODO: Return to UtilDrawGroundCoverQuad and get
it working properly.
5-May-2006
1.5
NOTE: To make an uninstall
shortcut, use the following as the 'target', replacing the product code with
the current one from the setup project:
C:\WINDOWS\system32\msiexec.exe /x {4C676BFE-488E-4A70-9DE4-C3E957723CD3}
Found you can get a 19"
(diagonal) 16:9 1440x900 thin monitor for not much more than £150.Tempting...
Tried to fix the dimple normal
generation.Decided I need to do this
by drawing the ground effect plane, not a screen-aligned quad.In this way, texture coords at any pixel on
the plane will indicate pixel's world-space XZ position.After offsetting that WS position for each
neighbour sample, need to multiply into clip space to do the height lookup
from ORT[3].
3-May-2006
0
Placed the order for my PC,
graphics card (X1900XTX!) and RAM.There's screen grabs of the order details, and order confirmation
emails, in the 'PC' folder.
25-Apr-2006
5
Got ground dimples and puddles
working pretty well, it's looking cool.
I'm a bit concerned about speed though - I think the bulk of the slowdown
is from the real-time normal-map generation (unsurprisingly!).
TODO: get dimple normal into proper plane space (in
PatchworkDimpleProcessing) before multiplying Ground's plane-space normal by
it.
TODO: Stop the dimple normals
getting so extreme in the distance.
23-Apr-2006
7.5
NOTE: Sundays are better for
using the office (emptier) than Saturdays.
NOTE: ATI SDK includes
RenderMonkey, CubeMapGen, and all the ATI shader papers.
MAX: Assigned ATI NormalMapperUI
to the N key.The tool seems to work
just fine, but what on earth is an NMF file?That's the only mesh format it accepts.
nVidia Melody accepts 3ds
files.It outputs normal maps in DDS
format, at the same location as the lo-res mesh.
NOTE: X850 supports OCCLUSION
QUERYING!See ATI
'OcclusionQuery' sample.
My CullPlane system isn't bad though; I'll stick with it for object
occlusion culling.
Occlusion querying might well be useful for occluding light flare effects.
ATI's Parallax Mapping sample
doesn't support POM on X850.Needs
SM3.
Ordered the sound card (M-Audio
Audiophile 2496) from Dolphin Music (http://www.dolphinmusic.co.uk).£60.99 total.->Had
to cancel this; first order has to ship to billing address.Would've been nice if they'd made that
clear before I ordered.
I'm getting a bit fed up of
using this under-capable graphics card (X850).If I'm going to be submitting this demo 3
years from now, it makes sense for me to be developing for the very latest
hardware I can get my hands on.If it
doesn't run on anyone else's machine for 3 years, so be it!
So when I get this 'rig' of mine, I'm going to fork out for an SM3 card
(likely ATI X1K series).Just seems
like a waste of time to be writing techniques that I'll end up replacing
nearer the time because they're too old-hat.
Decided I want patchy puddles.I don't want my ground to be as flooded as the ground in Toy Shop;
I want more variation.
Also I've noticed an absence of kerb-side streams in Toy Shop - I'll be
making sure those are present and correct in Noctambule.
Decided I want to add dimples to the ground.This is a coarse normal-map layer that
re-orientates the patchwork normals (currently the ground looks 'rough' but
still 'flat').
Also, puddle depth will be a function of dimple depth.
TODO: Draw dimple height to
ORT[3] RGB.Turn this into dimple
normals on ORT[2] RGB.
'Ground' shader combines patchwork normals, dimple normals and ripple
normals according to dimple height combined with patchwork height.
TODO: ORT[3] RGB: ripple
normals.
Removed MRT support from the
Patchwork system; it wasn't being used because it was somehow causing
problems on nVidia cards (even though the cards supported MRT).
TODO: Need a better system for
compiling the FX files.Having to add
each FX to the project and set up all its build properties individually is no
good.
Started adding dimple
decals.I can already see it's going
to make a big difference.
22-Apr-2006
5
Time to start on them puddles I
think.They're really important.
NOTE: The object matrix affects
verts' positions only, not their normals etc.This is good for ground patchwork decals because these decals get
flattened by an object matrix; this flattening does not affect their vert
normals (which can be used for fake undulation).
Had a proper look at the Toy
Shop demo & slideshow; read a lot of stuff about parallax occlusion
mapping etc.
NOTE: Apparently the POM method
described by Tatarchuk works on 2.0, 2.b and 3.0 in separate
implementations.Still haven't found
the code, but that would be mighty useful.<-must be in the ATI SDK (?) Seems to need SM3.
Found and downloaded the latest
ATI SDK.It's easy to find on the ATI
site (developer section, quick links).WIll try this out tomorrow.
Downloaded ATI CubeMapGen, which
generates specially neatened cube maps.Looks great.Removes the seam
artefacts between cube map faces.
21-Apr-2006
4.5
Scenics' CullPlane exclusion
codes now get propagated to all the scenic's descendants, using
CMeshFrame::SetCullPlaneExclusionsRecursive.Mosley was getting occluded by its own cullplanes.
Put default materials on the
trees to stop their reflections being pure white.Can't be bothered supporting un-materialed
objects; I'm happier using default materials.
Released build 4, the first
build to feature occlusion culling.
TODO: Find out why some of those
ground patchwork decals skew the reflection way off to the side.Are they just tilted in Max?<-Yeah, that was just something srewy with their transforms
in Max.Tried all the 'reset' type
options in Max but nothing helped so I'll just keep an eye out for them and
replace them from scratch as I find them.NB vertex normals in patchwork decal objects are honoured and should
be (better to put fine undulations into decal objects than into the ground
mesh).
Simplified the setup projects so
they now take their files straight from the VSS folders.
Also cleared out all unused files from Graphics\Final.
Made an instruction sheet for
the build process: Builds\howto.txt.
20-Apr-2006
3.166
Got CullPlane inter-exclusion
working nicely ^.^
Been testing with cpBTSouthEast and cpBTNorthEast (both now have
inter-exclusion set to 1).
This stuff is all in CCullPlaneManager::TestOcclusionVisibility.
19-Apr-2006
2.75
Ditched CullPlaneSuperchains
altogether.It's always a shame when
work goes to waste, but it could've been worse.
The CullPlaneChains seem to be doing a great job on their own anyway.
SourceSafe label: "about to ditch cp superchains".
Ground patchwork decals now use
occlusion culling :)
Placed more CullPlanes.
Started setting up a simple
inter-exclusion system for cullplanes, eg. for
cpBTSouthEast-cpBTNorthEast.
This is the 'chain curling behind viewpoint' problem case that I knew
existed but that I didn't think I would run into this quickly!
TODO: Return to TestOcclusionVisibility and
continue this.<-
done, seems to work really well!
So nice to be rolling again -
that superchain palaver was starting to get me down.It's vitally important to keep it rolling.
TODO: For objects that are
narrow on screen, do the occlusion culling using a single ray from a central
top point.
This will be super efficient because the object is occluded if ANY plane
covers that single point!
TODO: Eventually, should also
depth-sort the cullplanes - ideally very lazily, like once every couple of
seconds.
They would then be tested in near->far order which should drastically
cut down the number of ray-plane tests.
18-Apr-2006
2.17
Tried to think my way out of
this superchain problem, couldn't do it.
In case I should forget what the problem is at some point, here's an
example:
- look east along John St.
- cpBedfordNorthEast's corners are blocked by Mosley and Bedford
Court.
- Mosley and Bedford Court are in the same superchain, so
cpBedfordNorthEast is culled.
- cpBedfordNorthEast was visible in the gap between Mosley and Bedford
Court.
TODO: It's
a bit of a compromise, but I'm going to have to ditch superchains* and use
chains in their place (object is culled if its corners are covered by the
same plane or by two planes in the same chain).
This'll mean more ray tests and more objects passing the occlusion culling (only larger
objects thankfully).
But at least it should get me rolling again, which is always the most
important thing.<- Works pretty darn well
actually! :D
*Superchain occlusion culling
would work just fine for objects below a certain size though (eg.
characters).
But the smaller the object, the less gain there is from checking
superchains anyway, so I reckon it's best to ditch them
altogether.<-done
TODO: Place CullPlanes
everywhere.
TODO: Get tree reflections
properly coloured.<-using default materials.
TODO: Get ground decals culling
properly (not popping out).<-The
EMeshFrameFlag enum was out of sync with the
CMeshFrame::flags bitfield (so ground decals didn't know they were decals and
were using the wrong culling).
TODO: Release a new build.Compare framerate with v3.4?
9-Apr-2006
5.5
Had the minU and maxU swapped on
x-major cullplanes.Likes of
cpBedfordNorthEast is working again.
Eventually got the superchains
working just the way I wanted them, then realised the technique has a serious
flaw in it.
It's hard to describe though, and I'm not sure what to do about it
yet.
I'm amazed I didn't spot the problem when I was planning this out - it
should've been obvious...
8-Apr-2006
3.25
Installed the ATI Toy Shop demo,
but it doesn't run on my card.I'm
actually kinda glad!
The error message suggested that my card maybe didn't support the 1010102
format which is used heavily in the demo.
Made some progress with the
superchains.Really need to get more
hours in tomorrow.
Chose a PC I'm about to buy, so
thatI can work at home without the
noise distractions of the office (eg. other people's music).
This PC's got an onboard nVidia 6100 which should work nicely as a low-spec
test card.I'll add something a bit
niftier (SM3 I expect) to do the fancy stuff with.
I'll be getting a RAM upgrade to take it to 1.5GB, and a nice sound card.
30-Mar-2006
3
Got the CullPlanes exporting
accurately.I just had the direction X
component reversed, would you believe it!
Always check the obvious.
Can now freely scale
cullplanes.
TODO?: They still don't support transform resets
though.
JC reported a crash on his
nVidia Quadro4 900 XGL; that's a high-end SM3 card.
TODO: Look into this.Got a dxdiag
in C:\noctambule\JC_dxdiag.txt.
IMPORTANT: When 'Getting' stuff
recursively from SourceSafe to a different (eg. backup) location, tick the
'build tree' option on, otherwise the explicit working folders in sub-levels
will be used (dangerous!)
NOTE: PC375 does not have a DVD
writer!Only a CD writer.
Made a backup of the project.
TODO: CullPlaneSuperChains!Go go go!
29-Mar-2006
5.5
Re-installed the December 2005
DX SDK.This brought back the Xbox
surface viewer.
Fixed those link errors (see
above).
Max: Objects can take on a funny
scale when you connect them to a parent in the hierarchy.
CullPlane chains are now
successfully generated by the game on startup.
Improved CullPlane exporting a
bit.It's now definitely right as long
as you never use scale on them.The
two CullPlanes round the car-park wall are just temporary tests of this -
they can be discarded.
28-Mar-2006
3.5
Tried to get Max to export the
cullPlanes properly.Gave up because
it was taking too long and all the stuff that should have worked
didn't.
So from now on, I'm just going to have to make sure I
don't ever SCALE CullPlanes.
It's possible I'll have another go at it later in the project.
Reinstalling the Xbox XDK today
seems to have buggered my project.I
can't link now - I get stuff like this:
Mesh.obj : error LNK2001: unresolved external symbol
_IID_IDirect3DBaseTexture9
Reinstalling the Feb 2006 DX SDK didn't fix
it.I think it's the initial
underscore in the symbol name that's wrong.
TODO: Need to sort this out somehow - what fun.<- just needeed to delete (or demote) the XDK lib paths in
.NET\Tools\Options\Projects\VC++ Directories\Win32 - Library files.
TODO: Continue with cullplane
chain setup (CCullPlaneManager::PostParse)
17-Mar-2006
2
Made a doc in which to record
handy French vocabulary:C:\Noctambule\vss\docs\French
vocab.txt
Improved the PA logo &
icon.Worked the icon into the
installers.
I'm terrible once I start fiddling with icons and stuff - I can't stop!
Changed the installers to build
directly to website\final.
Removed debug configs to save confusion.
Saved Photoshop palettes (.act
files) for icons, in the Website folder.
16-Mar-2006
3
Installed Xara X.
Made a Programmer Art logo and
applied it to the game and website.No
idea why the transparency didn't work on the 16x16 256 colour version -
switched the 16x16 to 16 colours instead.
The logo is a P and an A; the P is meant to look like a set-square
(representing the programmer), and the A is meant to look like a warning
sign.It's like "Danger: coder
graphics ahead!"
Tried to get cullplanes exported
more accurately.Needs to take into
account the node's pivot transform or something, but I haven't quite worked
out how to do that yet.It's just
going to take a load more trial-and-error :/Needs to be done though.
15-Mar-2006
5.25
Website: Did a free submission
to 10 search engines.
Website: Set up email forwarding
- YAY!
This hasn't propogated yet ("Sender address rejected: Access
denied"); I'll keep an eye on it.
TODO: Work out why ground
patchwork decals are popping out.
Decided not to store cullplanes
grouped by chain, since some chains will be very long and contain planes that
will be small on screen.
Instead I'll allow for depth/distance/size-on-screen qsorting later -
should be a big CPU saving for the occlusion testing.
TODO: NWF Exporter:
Auto-generate and write-out cullplane chains.
TODO: Generate cullplane
superchains each frame (but not between eyes).
TODO: Take into account chains
and superchains in CCullPlaneManager::TestOcclusionVisibility.
The cullplane system should then be complete (basically).
Added the level .X and .NWF to
sourcesafe, just in case.
TODO: Return to CullPlane export
and get the planes written out accurately (see previous notes).
What's there at the moment can't be far off.<- given
up on this for now; workaround is to avoid using scale on them.
14-Mar-2006
4
Spent all evening trying to get
the landscape hierarchy pruning working.It was really very awkward, and I almost gave up.
It's working now though - currently managing to prune off 46 frames or so,
if I remember rightly, which is great.
TODO: NWF Exporter: Find out why
Mosley only excludes cpMosleySouth (should exclude east and north as well). <- was using 'exit' instead of 'continue', skipping all but
the first child plane.
TODO: Find out why CullPlane
occlusion doesn't seem to be working at the moment.
<- wasn't initialising cullPlaneExclusions to 0.
Max crashed on shutdown.Opened the level again quite happily;
didn't crash the next time.
TODO: sphere-frustum tests: use
a world-space frustum to save transforming the sphere into view space.
13-Mar-2006
4.25
Decided that CLandscape should
be what contains the (sorted by material/effect) scenic list and what
performs the scenic parsing.
C++: Memsetting a class instance
to 0 breaks everything if it inherits from another class.
The NWF file is now parsed twice
- once before the X file is loaded and once after.
Lights need to be parsed in before the materials in the X try to find
them;
Scenics need to be parsed after their frames have been loaded from the X.
I can't make the system any cleaner than this because I'm not in control of
what is and isn't written into the X file.
TODO: return to
CMeshFrame::PruneHierarchy.Get this
working so it removes all unnecessary frames, allowing duplicate visibility
tests to be eliminated.CullPlane
exclusion should then be working fully.
12-Mar-2006
3.75
Broke everything in the Max file
by trying to connect the whole landscape hierarchy to a new parent.This flattened the hierarchy.I had also forgotten that moving the parent
node moves all the child nodes with it unless you tick 'Don't Affect
Children' in the Hierarchy panel.
Fixed it all in the end by rolling back a few versions in SourceSafe and
repeating some work.
Max: Schematic view: When you
reconnect a node hierarchy to a new parent, it flattens the hierarchy!!
I expect I'll need to find a solution to this problem
later on - might need to write a script for it.
CullPlane exclusions now write
out to the nwf file.
TODO: Finish the code side of the exclusion
stuff.
Saw that ATI 'Toy Shop' demo on
Friday.It put me in a very bad
mood.I'm going to pretend it never
happened and carry on doing my thing.
9-Mar-2006
2.75
Found an online translator thing
at http://www.freetranslation.com - seems slightly better than Babelfish (but
still not reliable of course).
MaxScript: For plugin members
representing nodes, use type:#node rather than type:#maxObject.
The 'node' type gives you a proper node that you
can get the name of etc; I don't know how to get the name from a
'maxObject'.
This stuff doesn't seem to be documented properly.
MaxScript: using a class name
(even with different case) for a variable name is REALLY BAD!
For example if you assign a value to a variable called 'cullPlane', you're
actually redefining the CullPlane class as that value!
Scenics now export the name of
their designated CullPlane (the one that needs to be frontfacing for the
scenic to draw).
TODO: Return to
GetExcludedCullPlanes in NoctambuleWorldExport.ms.
Get the cullplane exclusions written out into the NWF.
5-Mar-2006
2
CullPlane exclusion: Planes will
not cull the following objects:
a. Objects from which the plane descends
b. Objects descending from the plane.
c. Siblings of the plane (and their children) - DEPENDENT ON A FLAG, TRUE
BY DEFAULT
d. Explicit visibility dependants of the plane (ie scenics that designate
it as a visibility controller)
Max: Found the 'always arrange'
button on the schematic view - switching this off makes it a lot more
useable.
Using 'arrange selected' where necessary seems nicer.
CullPlane: Added the 'Cull
Siblings' flag for case 'c' above.I
don't think I can automate things any further - sometimes you want planes to
cull their siblings and sometimes you don't.I'll keep a look-out for a pattern as I add the rest of the
planes.
'Top-level' type planes need to cull their siblings, while 'child' type
planes don't - but that's a vague statement because all this stuff can sit at
any level in the hierarchy...
TODO: Each frame carries a
bitfield indicating which planes it's NOT culled by.
When I run out of bits, maybe each frame can
start carrying one such bitfield per 'sector' or something like that...
1-Mar-2006
3.167
Installed ATI's 'fluid flow'
screensaver.Reminds me of the kind of
thing I'll want for my intro...
CullPlanes now test their own
visibility (once per frame in PrepareCullPlanesOnChangeOfView) to determine
if they should be used when testing occlusion of objects.
This is a fast but approximate plane-frustum cull.Planes that are deemed not visible are
drawn in red, so keep an eye on this.
Got the occlusion culling
working (more) nicely.It's so nice
seeing everything pop out when it goes behind the CullPlanes.
TODO: use this on ground patchwork decals too (precalculate their bounds
using ray tests down to the ground mesh).
Decided I'm going to stick with
occlusion planes.Having placed a bunch of them now I can see
that they're really quick and easy to use.On the engine side they're ideal too.
TODO: work out why the angle of
some of those CullPlanes ended up a bit skew-whiff.Mix of 'view' and 'local' rotation modes??
TODO?: On export, calculate the
plane based on the four vert positions (allowing complete freedom over
transforms during editing)?
TODO: Got to stop CullPlanes
occluding their 'own' buildings!
TODO: Fix current occlusion
problem cases.Make sure that the two
planes occluding the two corners overlap, that oughda do it.
Changed my devlog converter
thing so it doesn't crop the text at 1 meg's worth (now 8 megs, should keep
me going for a bit!).
28-Feb-2006
5.75
TODO: Remove the fly-cam from
the next build (it confuses people).
TODO: Tune the camera control
etc. to match Half-Life 2 or whichever other FPS is popular at the
moment.
Getting the game feeling right for people - early on - is more important
than I thought.
It does feel laggy/floaty at the moment, not sure why that is.
NOTE: Can't
use accents / cedillas in names of online files.If you do, you can't link to them or even
run them from the address bar of a browser.
Tested this in IE and Firefox.
NOTE: HTML code for 'ç' is
'%E7'.
Made a separate French installer
and uploaded that version to the site.Couldn't combine the two languages into one installer because I don't
have control over all the strings.
Uploaded a slightly-updated version of the English install too (tweaked
some strings etc).
Reinstalled DX Feb 2006 SDK,
because I was missing a dll and also I thought it might bring back my missing
Xbox surface viewer (it didn't).Didn't replace the missing dll either (so I fished it out of the bin
instead).
Max: Never use scale on
CullPlanes!It
gets too bizarre if you do.The
exporter now just writes out raw Plane properties - doesn't take into account
scale.
Got the basics of the occlusion
culling working nicely.Need to avoid
gaps where two planes meet though.Maybe they all need to cross over each other slightly at the
corners? - Yes they doMaybe I should be doing this thing using
lines (defining the tops of the occlusion 'walls') rather than planes?I'll have a think about it. - I'm going with planes, they're easy.
27-Feb-2006
5
Set up an installer using a .NET
'setup' project - gives me everything I need and more.
Currently just in English, TODO: French
version
(combine both languages into the same installer to save work?) <- can't.
I had to avoid using .NET SourceSafe
integration, because it was getting itself in a right twist.Just added the sln and vdproj files so
SourceSafe manually, on their own.
Uploaded v3.4 which uses this installer.
Setup: Default copyright warning
was
"WARNING: This computer program is protected by copyright law and
international treaties. Unauthorized duplication or distribution of this
program, or any portion of it, may result in severe civil or criminal
penalties, and will be prosecuted to the maximum extent possible under the
law."
Setup: Default update text
was
"Please use Windows Update to check for any critical updates to the
.NET Framework."
Setup: Banners seem to want to
be 500 x 69, even though only 497 x 69 pixels seem to get displayed
(?).
These details might well be different on different Windows setups though.
26-Feb-2006
6.5
.NET: Links in comments open the
file read-only :(
Restored visibility testing for
the ground patchwork decals.
Must have been switched off as a test.Need to use #pragma message to remind myself about that kind of thing.
MaxScript: Ran into a limit on
the number of 'else if's you can use together, in NoctambuleWorldExport.ms.
Set-up the game side of the
CullPlane system.It's basically all
in place but not quite working yet.
TODO: Continue this.
Shifted a bunch of maths
functions from Util.h to Maths.h.
Added some new maths functions.
25-Feb-2006
4.75
Decided the next thing to do was
to extend the landscape culling.
I'm going to go ahead with a system of visibility/occlusion planes,
beacause I think that's very suitable for my town environment.
Found Max's tree primitives in Geometry\AEC Extended, and placed a
few stand-in trees.They pretty the
place up a bit.
These have got an alpha pass that isn't currently drawn in-game; also their
reflections come out white in the ground reflection because they've got no
shader material on them.
TODO: fix this so they look presentable in the next build.
Max: Shifted my plugins into
dedicated Noctambule categories, eg. Noctambule Lights / Noctambule Geometry,
using the 'category' keyword in the plugin definitions.
My 'CullPlane' primitive wasn't showing up till I did this.
Added a 'CullPlane' plugin
(extends Plane).This is a finite
plane that must be visible and frontfacing in order for its dependent objects
to be visible; also it's an occlusion plane to block other objects' visibility.Should be very effective.
Added a 'Scenic' modifier (Object-space modifiers\Noctambule Scenic).
This will sit at the top of the stack on all
scenery objects.Currently just lets you specify a CullPlane for the object, using a
'pickbutton' control.
Started laying down
CullPlanes.
- These have to point their local Z axis away from the surface they run
along.
- The planes must be vertical - must only use rotation round the vertical world axis.
TODO: Change start point, spawn
points and collision rectangles to be plugin classes rather than using
user-property text.
CullPlanes now export from Max
in the NWF file.
TODO: CullPlanes on the game
side.
24-Feb-2006
0.5
M reported a big speedup between
v3.2 and v3.3 - presumably down to the major pruning of the bumpywall
buildings and the removal of the ground depth draw & test.
Went from 29/30 to about 40 according to Fraps.
Installed Firefox, checked that
the website works ok on it.Practically identical so far.I'll make sure I test new content on both browsers as I add it to the
site.
Asked IT for a copy of
InstallShield.
Labelled everything in
SourceSafe as 'v3.3'.
Thinking about landscape culling
in more detail.
23-Feb-2006
1.75
Got the patchwork / ground
effect working without MRTs - went surprisingly smoothly.Works on a 5900 now - YAY!
Uploaded v3.3.
Installed MSN Messenger.
22-Feb-2006
3.75
.NET: When editing web pages,
don't use the browse button to point a picture ref at a local image
file!!
The build 3 screenshot was missing from the front page because of this, and
I didn't even know about it because I had the local file it was poining at!
>:(
Trying to fix this reflection
glitch on the nVidia cards.I'm using
T's 5900, which is a handy minimum-spec type card for me (the game runs at a
snail's pace on that machine!).
- Tried taking the ground depth clipping out altogether.Made no difference.Note that the zombies' reflections are
chopped up just the same as the building reflections.Interesting that on this 5900 the
reflection is mangled out of position too, not just gaps like on the 6600
& 6800.Taking the ground depth
clipping out did get rid of the rendertarget contention warning I was
getting.
- Reflection image is corrupted, but looks like it's mapping correctly onto
the ground.
Been getting this from
CPatchwork::Render where it draws the receiver:
"Direct3D9: (ERROR) :DrawIndexedPrimitive failed.
D3D9 Helper: IDirect3DDevice9::DrawIndexedPrimitive failed: E_FAIL"
Took the size of the reflection
ORT down to 512 - this has a lovely softening effect on the reflections.Prevents fizziness in reflections of
brickwork etc.
Added an un-set of
rendertarget[1] after the patchwork draws!
Started seeing this (when
drawing general landscape) when I enabled 'maximum validation' in the DX
control panel thing:
"D3D9 Helper: IDirect3DDevice9::DrawIndexedPrimitive failed:
E_FAIL
Direct3D9: (ERROR) :Vertex shader function usage (D3DDECLUSAGE_TEXCOORD,0)
does not have corresponding usage in the current vertex declaration
Direct3D9: (INFO) :The vertex declaration is (Stream, Offset, Type, Method,
Usage, UsageIndex):
Direct3D9: (INFO) :0, 0, D3DDECLTYPE_FLOAT3, D3DDECLMETHOD_DEFAULT,
D3DDECLUSAGE_POSITION, 0
Direct3D9: (INFO) :0, 12, D3DDECLTYPE_FLOAT3, D3DDECLMETHOD_DEFAULT,
D3DDECLUSAGE_NORMAL, 0
Direct3D9: (INFO) :0, 24, D3DDECLTYPE_FLOAT3, D3DDECLMETHOD_DEFAULT,
D3DDECLUSAGE_TANGENT, 0
Direct3D9: (INFO) :0, 36, D3DDECLTYPE_FLOAT3, D3DDECLMETHOD_DEFAULT,
D3DDECLUSAGE_BINORMAL, 0"
NOTE: 5900 can only render to
one RT at a time :(
Tried to get a non-MRT version
of the patchwork shader working.Couldn't get it to behave.Keeps ignoring the values I set for 'twoPasses' and 'currentPass' (or
maybe doesn't test them properly, who knows).
Maybe try disabling preshaders as a test?
TODO: Get this working!
20-Feb-2006
7
It seems that Max will always
crash on trying to load a level that uses a shader that doesn't compile.
Max has a nasty habit of caching
textures so that when you update the texture file the change doesn't make it
through to Max.
Restarting Max fixes this.
Max: When adding any objects,
remember to put them in their right place in the hierarchy!
TODO: Ground effect plane needs
to lerp around discreetly - currently you can see it snapping when you step
off a kerb.
Because of the space in the name
of the folder "Batch Files", I had to add quotes to a path using
that folder name, in build.bat
AnimUser.cpp: Moved
'AdvanceTime' call from render to update.This stopped the zombies animating at double speed.
Added a bodge to correct the
zombies' walk speed.Noticed they're
all walking in step at the moment - is that new?
M&G's 6600 GT is 128 meg.
Had to change the run.bat to
copy d3dx9_29.dll into C:\windows\system32.Need to find a better fix because I can't
rely on that path.
Uploaded build 3, then build 3.1
with the above fix.
RenderMonkey's 'package
exporter' isn't what I thought it was at all.It just makes a zip containing the RFX file, the meshes and the
textures.Very handy if you want to
share your shaders with other people, but I was looking for the exact
opposite!
Relased build 3.2, which uses
pre-compiled effects (fxo files) so that no-one can see my code.
19-Feb-2006
11
Max: NOTE: The rendertarget
scripts in patchwork.fx were causing max to crash when loading leam.max.
Max: M for material editor.
Max: Object render order is not
depth-based :(
I've now changed the exporter so it 'ifndef _3DSMAX's zwrite=false and
ztest=always.
Also changed it so it removes the scripts that max doesn't like (removes
them even for ingame, since I don't use them).
patchwork.fx: fixed the
transformation of the normals, pretty much by guesswork.
HLSL: A PS float3 normalize uses
three arithmetic slots.
TODO: (after build 3) remove
CMesh::m_pTextures and m_pNormalMaps!!And any other out of
date junk.
Swtiched back to DX debug libs,
and started seeing this:
"Direct3D9: (WARN) :Can not render to a render target that is also
used as a texture. A render target was detected as bound, but couldn't detect
if texture was actually used in rendering."
I don't actually think it's a concern.
Had to remove 'd3dx9dt.lib' from
the Release and Master configs otherwise they crashed.
Reconstructed my 'release'
config because I was unclear about why it wasn't working properly.
After doing this I got a link error about 'test\winmain.res' not being
found ('test' was the original name of the new release config).This went away after I restarted .NET.
Got a hefty speedup from
minimizing the distance that the buildings extend below the ground.
Got a hefty speedup from drawing
the landscape with CCW culling normally and CW culling when reflected.
Got point lights working on the
zombies.Just had to use the device's
LightEnable method after SetLight.Didn't need this for a directional light.
TODO: Get proper memory leak
reports working.
Trying to track down those
leaks.
- There's 134, that's 2 per CMesh instance.
- Not in CMesh::SetVertexDecl
- Not it CMesh::GenerateSurfaceVectors
- Nothing to do with CMesh::m_strMaterials
- Not in D3DXComputeTangentFrameEx
- Nothing to do with OptimizeInplace
- Nothing to do with InvalidateDeviceObjects
- " D3DXComputeNormals
<-Fixed it - pEffectInstanceBuffer in
CMesh::Create
Reenabled CMesh frustum culling.
Discovered a last-minute ground
reflection bug.Spent quite a long
time fixing it.
Everything is working properly
now, at last.Build 3 will go online
tonight (Monday night).I could do it
now but I'd be here till 3am, which wouldn't be so nice, and I'd rather not
rush it anyway.
TODO: Straight after build
3
- back up the project!!
- get the FX code hidden (point release?) v3.2
- improve the devlog convertor thingy so it lets
me hide rows (eg. those that describe effects processes).Replace all non-space characters with '/'
or something.
TODO: for build 3: normal map
for yellow road lines.
18-Feb-2006
11
Got the '__Specular' and
'__Diffuse' properties of bumpywall editable in Max again.The exporter had just been culling them
because they weren't used.
MAX: Couldn't find a way of
getting the mesh's original vertex colours through to the shaders.That would be really handy.
TODO: Need to work this out.
7.5.6: The shader's stream mapping needs to
include a colour element!
MAX: Found something to do with
DX light maps in the material editor.TODO: investigate.
Max's 'subdivision' modifier is
my new best friend.
Added normal mapping to the
reflective ground shader - SO PRETTY!
COOL!RenderMonkey does support multiple
simultaneous render targets!It's
absurdly easy to set up too.
I've applied this to the 'DummyPatchworkImage' pass of the 'ground' shader.
TODO: Render the patchwork
decals like so:
- Make a 'patchwork' shader for the decals, allowing
them to write their rgb and their normals+gloss to two simultaneous ORTs.
- Use the patchwork shader to make materials
for the decals (one per texture)
- Each material specifies its colour map and
normal map
- Make a 'puddle' shader & materials for the puddle decals.
- The puddle decals will, initially, just ADD gloss into the alpha channel
of the normal-map ORT.
MAX: NOTE: The 'Instance
duplicate map' menu item in the material editor can cause a crash when you
click it.
Did a half-decent job of making
a height map for that manhole cover:
- trace contour
- Gaussian blur
- PSP dilate
- curves
- Multiply layer
Confirmed that an 8-bit PNG is
smaller than an 8-bit BMP (though only a tiny bit smaller in my test case)
Photoshop: Confirmed that you don't need to set number of colours
to 255 when reducing
to 8 bit.Just make sure the forced
colours option is set to none.
I've been using 'local adaptive' for normal maps and 'local perceptual' for
colour maps.
TODO: FX exporter: Remove render
target scripts because Max doesn't like them.
TODO: Some of those manhole
covers have still got their surface vectors pointing the wrong way.
17-Feb-2006
1.5
NOTE: X850 and 6200 both support
up to 4 simultaneous render targets.
Started naming HLSL
functions/variables/structs like C++ ones.
TODO:Render planar patchwork to two
simultaneous ORTs
ORT1:(R,G,B,[Height]) - will need new shader
for the water influence on height.
This height is final (it includes distortion by rain impacts and puddle
surfaces)
ORT2:(TsNx, TsNy, TsNz, [Gloss inflenced by
water and by tarmac height but not by normals]) - will need new shader for
water influence on normals & gloss.
These tangent-space normals are final (they include distortion by rain
impacts and puddle flow).
--
Then draw the ground mesh to two ORTs.The height data from the patchwork ORT can be used for parallax
displacement at this stage.
ORT3:raw R, raw G, raw B, [Gloss
influenced by normals]
ORT4:Nx, Ny, Nz
--
Then draw the ground + lighting & reflection in a single screen-space
deferred shading pass.
16-Feb-2006
5.5
RM: Don't need to save before
exporting.
Got the ground drawing again -
in CMesh::Render I had been nulling the VS & PS if the mesh's effect
pointer was NULL (this was stopping the ground effect doing its thing).
FX exporter: TODO: Stop the auto
duplicate-remout applying to local variables!
Added a common shader effect
bool param "groundDepthClipping".Used to clip reflection geometry to the ground depth (borrowed from
GroundReflection.fx "ReflectionGeometry" pass).
NOTE: RM can crash suddenly when
you're typing something.Save
frequently; don't trust it.
Disabled Max's crash report
thingy by renaming it 'DONT_SendDmp.exe'
Found that Max totally can't
handle #includes in FX files.I can see this being a bit of a problem
later on - maybe I'll have to make the exporter expand the includes into the
exported file.
TODO: Do this, so I can start using FX include
files for structs and stuff.
Stopped using the
'ReflectionGeometry' pass of groundReflection.fx altogether.
Ran into a PS arithmetic
instruction slot limit (for the first time) in bumpywall.fx.It was just test stuff clogging it up,
nothing to worry about.Did have to
knock out luxury normalizes though, but I need to improve the modelling
anyway.
Got the game pretty much back
into one piece, and it's starting to look nice.Build #3 should be ready this weekend if
nothing else gets in the way.
TODO for build 3:
- Improve zombies' placeholder lighting (use
same light as bumpywall).This
includes their reflections.
- Improve building meshes to minimize lighting
artefacts etc.Improve texturing too.
- Are the buildings being culled at the
moment?If not get that working (if it
looks like a big job leave till after build 3 though)
- Normal maps for the ground
- Reduce ground reflection distortion to give
the impression of perspective
- Blatantly-placeholder lightsource effect for
the bumpywall light (yellow sphere - simple colouring shader)
- Fix shutdown memory leaks
- Polish bumpywall (most importantly, include
diffuse lighting element)
- Pre-compile all shaders so the embarrassing
source isn't visible
- Get master build running
- Make fly cam available in master build
15-Feb-2006
1.75
Landscape: Had to use D3DXGetDeclVertexSize in place of
D3DXGetFVFVertexSize, because
I'm not using FVF-code-compatible vertex declarations on the meshes any more
(so ID3DXBaseMesh::GetFVF was returning 0).
Started getting this message on
exit:
"D3DX: MEMORY LEAKS DETECTED: 134 allocations unfreed (16169
bytes)
D3DX: Set HKLM\Software\Microsoft\Direct3D\D3DXBreakOnAllocId=0x570f to
debug"
<-Fixed it - pEffectInstanceBuffer in CMesh::Create
I've disabled general landscape
draw for the moment while I get the ground working.Have I got a transposed matrix somewhere??
14-Feb-2006
4
Got BumpyWall working perfectly
in-game :D
I just needed to wait till after I'd pre-transformed the meshes into world
space, before calculating their surface vectors.
Still got some artefacts but they're due to bad modelling, not shader problems
(eg. see output from CMesh::GenerateSurfaceVectors).
In Max, tangent-space normal is
based on bumpCol.yxz; in RM and in game it's xyz.Not sure why it's different in Max.
HLSL: Multiple #defines of the
same symbol are a-ok.
FX exporter now #defines '_3DSMAX' at the top of every file,
but wraps it in '#ifndef INGAME'.
Don't believe any talk of a built-in _3DSMAX_ define, there's no sign of it
when using Max 6.
Now using normals from the X
file again.I'm pre-normalizing them
and pre-rotating them into world space at startup.
Tried removing as many
normalizes as possible from bumpywall.
- removed all from the VS, ingame (I'm pre-normalizing normals at
startup)
- TODO: remove per-pixel norms of incoming surface vectors (this brings
down the quality of the effect a bit, but better modelling should help)
- removed norms of reflected vectors and rotated vectors
- can't remove norms of the normal map colours, unless filtering is all
POINT and all normal-map mip levels have normalised texel colours (TODO? for
distant surfaces?)
Confirmed that the nVidia
normal-map plugin for Photoshop generates normalised texel RGBs.
In master builds, shader effects
are now compiled with the same '_MASTER' define.
Experimented with the sampler
states on bumpywall.As we might
expect, maxed-out anisotropic filtering all the way gives the cleanest
sharpest results.Currently set to 8
levels.I expect the 6600GT can do 16
same as mine.DX9 lowest common
denominator is 4.It looks like
requests for unsupported numbers of levels causes it to use what looks like
linear filtering instead.
Tamed the moiring (especially on the normal map sampler) using mip
settings.
TODO: Get the ground back!!
13-Feb-2006
3
Installed February 2006 DX SDK
and its debug symbols.
Confirmed that the effect Set...
methods do take a COPY of the data you pass to them.
I think the viewpoint Y
discrepancy was nothing at all.The
-1170 height was just the one used when drawing reflected buildings (1170-830 = (170*2)).I've disabled the reflected draw for just
now.
Seems I've been a bit confused
about these shader matrices.
Regardless of whether the matrix is declared row_major or column_major:
_41,_42,_43 is always translation if the matrix uses a '...Transpose'
semantic in RenderMonkey.
When the matrix is column-major, the watch-window display shows translation
in the 4th column; when it's row-major the matrix doesn't display in the
watch window at all.
NOTE: Setting an effect instance
vector parameter in SetMatrices will have no effect because it will be
overwritten in the Apply method with the value that the effect instance has
stored.
Found that the PS normals are
all wrong on Bedford Court (the ones on the car park wall look ok though).
Found that in Max, the viewpoint
is always at the origin and it's the world that moves and rotates.
Forced new normals to be
generated for meshes, but it made no difference - Bedford Court was still
wrong in exactly the same way.
Max got 'stuck' with bumpywall
again.Tried the usual tricks but
couldn't get it to behave.
Positioned a light right up by
the car park wall and got a result that's almost right.You can definitely 'see' the light
source.Looks like the binormals and
tangents are just twisted round 90°, or something like that (the peripheral
highlights are on the wrong edges of the bricks).I still don't know why Bedford Court's
normals are so weird.
Need to call it a night there,
otherwise I know I'll have a load of bad dreams about twisted surface vectors
:/
12-Feb-2006
3.25
HLSL: NOTE: Matrices are
column-major by default, and apparently this tends to be more efficient in
terms of instruction slots used when multiplying vectors by them, etc.
Row-major matrices can't be viewed in a watch window.
Started adding my FX files into
sourcesafe, because sometimes I edit them directly.
bumpywall: confirmed that
viewToExterior._41,_42,_43 is the viewpoint.
Confirmed that the binormals and tangents in rendermonkey are ok.
Fixed a stupid bug: pixelToVP_unit was being calculated as viewpos-lightpos
or something!The effect is looking
right in RM again now.
Made more progress towards
getting bumpywall.fx working in game the same as in RenderMonkey and
Max...
NOTE: Currently using the viewpoint as the light source for this (test).
Added WASD controls, in response
to customer feedback ;)
10-Feb-2006
0
Something about that
viewToExterior matrix getting transposed and also not getting the lesser-used
column/row initialised...
9-Feb-2006
3
Note: There's surface viewers
available in .NET, in Debug\Direct3D.
Got shader debugging working
RARR!
Made more progress towards
getting bumpywall.fx working in game the same as in RenderMonkey and Max...
NOTE: Panda exporter exports
normals but not binormals or tangents.The official max 7 exporter says it does export them.
For release build, had to remove
the flags D3DXSHADER_NO_PRESHADER | D3DXFX_NOT_CLONEABLE from
CEffect::Create.
Otherwise got the message "D3DX: Flags parameter is invalid" and
a break.
Next, got the error "D3DX:
ID3DXEffect::SetArrayRange: Invalid handle" and a break.Unhelpful call stack.
TODO: Fix this!
Got rid of the flickering in
bumpywall - viewToExterior was undeclared in the PS.
Bumpywall is no longer changing erroneously when the view orientation is
changed.
Now amending the the meshes'
vertex declarations from the X file to include calculated binormals and
tangents.
There should be no ambiguity in calculating these since they should just
align with the texture mapping, and there's already exported normals.
Bumpywall is still wrong somehow.
Tried auto-computing normals as well but this
made no difference at all (which is maybe a good sign). <- Normal generation was being waived because normals were
already present.
Found a Y discrepancy between
g_viewInfo.position (eg {-10942.8, -830, 8794.94}) and the 'viewpoint'
variable in the bumpywall PS (eg { -10942.8, -1170, 8794.94 }).TODO: Fix this!
8-Feb-2006
4
Eventually got to the bottom of
the 'data size mismatch' error when applying effect instances:
Max exports combo-style effect parameter values as
dword combo row indices (zero-based)!
So my 'PointLight01' value was just exporting as
a dword '2', rather than the position vector of PointLight01.
Changed the PointLight plugin so
it exports each light's auto-calculated 'shader index' (see above).
Changed the ordering inside
CModeEnvironment::OnAddedToStack, so that effect instances parsed from the
level .X file can lookup point lights by index.
NOTE: FX files currently have to
be in Graphics\Final for the game.In Max, I'm currently using the copies from
source\HLSL.TODO: Sort this out.
Now always using source\HLSL.
TODO?: Texture most landscape
mesh in 1 metre tiles, then use mapping density parameters in the
shaders/materials?
That makes it loads easier to change an object's texture without having to
update its mapping.
Made progress towards getting
bumpywall.fx (my test shader) working on the landscape objects in-game.
Still not quite working yet, but it should be close.
TODO: See if the meshes in-game
have binormal & tangent data.There's wacky numbers coming from somewhere; maybe it's just those
vectors not being set up at all.
Right enough, they were all 0,0,0.
28-Jan-2006
6
Had to think about how I'm going
to store and render the landscape.
- Keep the hierarchical mesh layout because it's great for culling.
- Divide the landscape into visibility 'sectors' (just special
frames).Need to divide the ground
according to these sectors too, so maybe the ground frames can be the
sectors' top-level frames (Max will need an object of some kind to create a
frame, and I'd rather not use dummy bounding boxes).
- Within each sector, have a second layout that sorts mesh subsets by
material, then by effect.
- Deal with alphas later (trees, fencing, windows etc).Ignore all that for just now.
Set up appropriate culling for individual building faces later.Each face will be a separate frame, a child
of the building frame - will only draw if you're on the right side of its
plane.I expect all the most expensive
detail will be in the building fronts (shop interiors, etc).
- Cull shop interiors based on visibility
(and size-on-screen) of the windows.Windows need to tint-over when small to hide the interiors.This will seem natural in many cases
because windows farther down a street will tend to be viewed from a shallower
angle anyway.
- Need a system whereby meshes can be re-used by multiple instances of an
object.Hmmm.Look at the way the statemanager sample
uses FrameMeshName blocks in its scene.x.Can Max be made to export that kind of block?Look into reference objects and X-refs,
whatever they are.
Excel is rubbish at dealing with
lots of text in its cells, but you can sometimes help it along a bit by
adding some newlines before/after the text.You can also make several rows look like the same one by knocking borders
off (as above).Need to make sure you
don't start the line with '-' though.
TODO: Make use of that nice
CResource class that came from the StateManager sample.I haven't used it for CMesh though.
NOTE: For shader stuff, object
matrices seem to be referred to as 'world' matrices.So object space seems to be referred to as
'world' space, how confusing.
My 'bumpywall' effect, the
version that worked properly, seems to have been erased from history.
I don't understand.I was saving, I
was checking-in, saving, checking-in.But it's not in the current RFX and It's not in any of the sourcesafe
history versions.HOW??!!Maybe I started to get slack and use F7
instead of saving?Looks like I didn't
do a checkin at the very end of the 26th when I got it working.I do remember getting an error message
(that I ignored) from SourceSafe about a 'delayed write' failing - could that
have anything to do with it?
TODO: Verify important code checkins!
I'm going to re-write that effect right now, even if it takes all night!!
>:(
TODO: delete all the really old
backups in the Noctambule folder; back up the project again.
TODO: FX exporter: Make sure it
handles auto-remout of 'shared' and 'row_major' variables!
NOTE: In RenderMonkey, F7 DOES
NOT SAVE.It removes the asterisks, to
make you *think* it's saved, but it DOES NOT SAVE.
Also, SourceSafe waives checkin
altogether if the file hasn't changed, which made the history record a bit
confusing.So what I must have been
doing is using F7 instead of saving - then for some reason not saved before
exiting - then exported over the working FX file earlier tonight.
RIght, I've re-written
bumpywall.fx.Luckily it took no time
at all the second time round.
I've checked it in and verified the check-in this time!Phew!
Tried to work out what the 'data
size mismatch' problem was all about; didn't quite manage.
TODO: Need to fix this!
It's like the materials' effect instances have
got parameters for the wrong effects.
27-Jan-2006
2.5
Seems the official Max X file
exporter only works for Max 7.
Installed the Panda one instead.
I don't have a copy of the Max 8
SDK.Apparently you need to be a
'registered developer' and pay a fee to get this.It's also part of the Max 8 DISK install
(my version is the download version).
Emptied my recycle bin, which
seems to have got rid of the ten-minute wait whenever I delete something.
Wowza - Max 8 comes with a few
more example shader materials.
Hmm, can't export anything using
my shader materials from Max 8.The
exporter crashes.
I can export diffusebump, glow, all the basic
ones.
Textured_bump crashes it.
Well I could spend hours working out what exactly it doesn't like, but I
think it'd be a waste of time.
Since it doesn't seem to have an option to export Max 6 files, Max 8 is
pretty useless to me for the mean time.
TODO: Come back to it later, see if the exporter's been improved / work out
what's crashing it.
NOTE: looks like 1x1 textures
are a-ok.'EffectParam' sample uses
one.
Started setting up an effect
pool and effect param blocks.
26-Jan-2006
4.75
Installed Max 8 (an, erm, "
evaluation " copy from G ;)
Had some buildings go flickery
when I added a shader material to them (like it was drawing an unshaded
version under it).
Was the same on Max 6.
Converting to editable mesh fixed this.
Ooh, there's asset-tracking
stuff in Max.Is that new?
Put a copy of leam.max in the
'safe' folder before saving a max 8 version.
Set up keyboard shortcuts again
- F for pan camera mode and D for fov.
Saved keyboard config in the noctambule folder ('kbd' file).
Looks like the DX SDK only comes
in English.Installed the Dec 2005
version.
Kept a safe copy of the official
max .x exporter that I tried to get working for max 6:
C:\Documents and Settings\PPalmer\My Documents\Visual Studio Projects\Max
(TAMPERED)
NOTE: FX parameters bit in
material editor sometimes needs a bit of help updating after the effect's
parameter semantics have changed.
MAN you've got to watch those
W's when writing shaders.DON'T MIX
FLOAT4s AND FLOAT3s IF YOU CAN AVOID IT.USE FLOAT3s WHEREVER POSSIBLE!!!!!!!!!!!
The new & improved bumpywall.fx was buggered for hours because the
normals, tangents & binormals were coming in to the VS as float4s but
then I was using float3s the rest of the way.I suppose it would've been the 'mul's that were going crazy?
Spent most of the night changing
bumpywall.fx so that it hopefully wouldn't have those Gouraud
artefacts.
And in the end, eventually, it worked!It works stunningly!I can
barely believe how perfect and promising it looks.
I was misled for a long time by Max not reshoogling Bedford Court in
whatever way it needed to - I didn't think I'd managed to fix the
artefacts.But I put a subdivide
modifier on the building, then took that modifier off, then the building was
properly reshoogled.It even fixed
some dodgy normals I had (?)
Old version is
BumpyWall_perVertex.fx.
YAY!Max has finally got a fly cam! :D Needs a
bit of taming though - it's way too fast at the moment.
Called "Walk Through" camera.
25-Jan-2006
4.33
pos & normal both in
object's LOCAL space.
World-to-view converts it properly into view space though.
Light pos in WORLD space.
World matrix is being filled in as rotation/scale only!!!!!!!
Max recognises:
ViewInverseTranspose
View
ViewInverse
…
NOTE: RM currently doesn't
export semantics for unused matrix variables!!
Max doesn't fill in a 'View'
matrix translation.
Max does not recognise:
ViewTranspose
Projection
ViewProjection
…
RM FX exporter now strips off
'Transpose' from all semantic strings.TODO: Conversely, add 'Transpose' where needed.
Right, FINALLY got the shader
materials working properly with the light positions AND the object
transforms.
This took all night - I had to work out which matrix semantics were being
recognised by the Max plugin, which ones it wasn't recognising, which ones it
was mis-interpreting and how.
In Max, vert positions arrive in the VS in the
object's LOCAL SPACE!!!Took me some time to work that out.
I'm therefore having to transform vert positions, normals etc into view
space then out into proper world space.
TODO: See if the shader
materials can be made to work properly on sub-objects (currently uses some
wrong transforms somewhere along the line, but to be honest, I can live with
it if I have to).
(No problem there - it was just that some materials had
been set to use the wrong light).
Should receive a nice present
tomorrow that might make things easier…
TODO: Install December DX SDK,
then get this bump effect through to the game.
24-Jan-2006
6
Got sourcesafe integration
working on the RM FX exporter and the devlog converter.I hate sourcesafe.
IMPORTANT: Max shader previews
TOTALLY DON'T WORK in orthographic views!!
NOTE: Have to normalize normals
for Max.TODO: wrap this in
"#ifndef INGAME".
A 'normal' modifier with 'unify normals' ticked doesn't help (but 'flip
normals' does - weird).
CAUTION: HLSL lets you do
'dot(vec1.x, vec2)'!!!
This wasted so much time :(
TODO: Look into ID3DXEffectPool.
NOTE: Max doesn't seem to be
case-sensitive about its variable semantic names.
So relieved it recognises 'WorldViewInverse' (view->world, so I can get
the viewpoint easily).
Made a specular bump effect and
got it working properly(ish) with the light positions in Max.
Took a lot of faffing and hair-tearing to get this going (much more than it
should have), but I learned plenty from it.
Max doesn't take into account objects' local transformations, so currently
I've just got a single light near the origin, working more like a
directional.
TODO: Get rid of the Gouraud
artefacts in bumpywall.fx - shift some calcs from VS to PS.
Learned about tangent space -
all far more straightforward than I'd imagined.
23-Jan-2006
4.75
Wrote a little program to format
this dev-log for the website (put most recent day entry at top, etc.)
NOTE: Excel is flakey at working out row heights and this transfers to the
online log; I might do something about this later but till then we'll have to
put up with random missing text.Only
seems to be a problem when the cell has a lot of text in it.
Moved 'Acknowledgements and
Resources' into a separate Excel file to facilitate html export of this log
(exporting page only loses a bunch of formatting).
Had to hide some dev-log cells
so they're not present in the online version.String "<hash>hide" (case sensitive), anywhere in the
day entry (eg. In the date row), denotes a hidden day entry.TODO: better filtering system.
I'm going to keep the source log
in html format since Excel seems to know what it's doing with it.
Started putting copies of the
devlog on the site.
22-Jan-2006
7.66
M&G's graphics card is an
nVidia 6600 GT.Should be able to do
anything my X850 XT can do.
TODO: Get some InstallShield
action on the go.
NOTE: FX FILES: First int of
MeshMaterialList block is a number of materials...
TODO: See if Max understands the
'VIEWINV' semantic?
Started cannibalising the D3D
'State Viewer' sample for my landscape.This has just started working - got the effects coming through and
looking just like they do in Max!:D:D Got a few parameters to hook up though,
but nothing too taxing I'm sure.
NOTE: Debug output: "D3DX: ID3DXEffect::SetValue:
Data size mismatch".This is what's causing the chugging.
All the old landscape code is still in place - TODO remove.
For future reference, this was a very good code-upgrade method.Leave all the existing code in place and
just append the new code to the end of the appropriate functions (denote the
new code sections with comments of course).
21-Jan-2006
9.25
NOTE: Using D3DXMesh, model
parts with different effects need to be separate meshes.
But within a single mesh, multiple different Max materials can be used, and
these can send their paramters (including textures) to the effect.Material = effect configuration.
So, I'll specify effects at the
CMeshFrame level.Effects will be
inherited down the frame hierarchy.
I wish I had my own Max
exporter.
TODO?: OOD?: For the mesh
system, make a 'material' wrapper that wraps the d3dx material with an array
of textures that will all be sent to the active effect before the mesh
draws.
See D3DXMATERIAL.
Now using CMaterial from the StateManager sample.
No way!There'sMaxScript menu in max, with a 'visual maxscript editor' in it!When did that happen?
WICKED!!You can create Max materials of type 'DX9
Effect' and it gives you a gui for all the effects parameters!And it understands the Max lights as
lightsources for the effects!And you
can see the effect in real time in the viewport!And the effect info gets exported to the .x
file!!
The only problem I've seen so far is it forgets which effect 'technique'
each material uses (I'm guessing it's defaulting to the first one).
NOTE: sometimes nvidia normal
map generation does weird stuff if you ask it to put height into the alpha
channel.Maybe best to do this
manually.
TODO: Change the RM geometry
loader so it game-aligns meshes coming in from Max.
Spent ages adapting the RM FX
exporter so it pipes the effects nicely into Max for use in materials.
Works a treat now.
Started chucking shiny bumpy materials onto objects all over the
place!So cool!
TODO: Get the new effect
materials through to the game!
TODO: Get the normal mapping on the ground working!
NOTE: BumpyWall.fx isn't using
lightsource positions properly in Max.The default light source (viewpoint) currenty works best, but even
that looks like it's slightly off.This is only a stand-in effect anyway…
20-Jan-2006
5.5
It seems that Max 7 has very
good normal-mapping tools/support.
nVidia have got an impressive
selection of shader-related tools on their site.
Installed an nVidia
heightmap-to-normalmap filter for Photoshop.
Installed nVidia 'Melody' which might be handy when I start modelling
objects.
TODO:Render patchwork to multiple simultaneous
ort's (no more than 4).
ORT1:R, G, B, Height
ORT2:Nx, Ny, Nz, PuddleFlow
ORT3:…
PUDDLES:
- Puddle depth is based on overall height image combined with vertex
normals (steeper ground is less puddly).Ideally puddles would slump sideways depending on ground normal, but I
can't think of a nice way of doing it at the moment.In addition to the height maps on all the
patchwork decals, I'll also have a height-only coarse noisy bump layer which
will be the main influence on where the puddles appear.
- Puddle direction is based on vertex normals, times the puddle flow
multiplier channel.
- Flowing puddles will be done as flow-only patchwork decals lining the
kerbsides.I suppose the flow
intensity should imply puddle depth too.
Draw
general patchwork with RGBA textures on 0 and NNNH textures on 1; write RGBH
to ORT1 and NNNX to ORT2.
Draw flow patchwork with F textures on 0; write XXXF to ORT2.
Draw moisture patchwork with
NOTE: Pixel shaders can output a
DEPTH value!
Installed FX Composer.Certainly has its merits, but having to
type in all the FX code makes it a no-go for me.
The "kick-ass" relief-mapping demo doesn't work on my card (it's
2.a and my card only goes up to 2.0).
Looked through the examples.Couldn't see anything mindblowing.The glass sample is nice.
Gathered some reference photos
from t'internet of wet streets at night.
TODO: Decide what to do about
ground relief (and relief in general).
Started automating the gathering
of bump map textures etc as the meshes' textures are loaded.Needs improving…
19-Jan-2006
3.66
NOTE: In RM, don't seem to be
able to update variables in one shader to be picked up in a later one.
Eg., can't update a matrix in the VS and use those values in the PS.
IMPORTANT: In HLSL, before
normalizing a four-element vector, you have to make sure its W component is
ZERO!
I have no idea why.
Finally got the ground
reflection distorting properly according to the surface normals.
From now on, I'm going to use
game-aligned test models in RenderMonkey - does away with that axis
conversion crap entirely.
Got the ground shaders working this way.
NOTE: the ground reflection
disortion is only "correct" for a vertex lying on the line of
sight.This is not a bug.
Subdivided Bedford street a bit
because some of the funny reflections had been caused by huge / splintery
polys.
TODO: Tidy up the existing
shaders, the way they're all set up and that (game and RM).
TODO: Decrease the ground
reflection distortion into the distance, to account for perspective.NOTE this makes sense for buildings/lights
but not for the moon (far less important).
TODO: Incorporate normal maps
into the ground reflection distortion.
TODO: Get some prototype puddles
in, then put build #3 on the site!
18-Jan-2006
6
Got my zombie reflections
drawing again.It was just out-of-date
ORT clearing resulting from changes to the reflection shader effect.
TODO: Try using Subversion for
the web project instead of VSS (which doesn't seem to be up to the job).
Returned to the single
Nocatambule.rfx shader project for the whole game - MUCH nicer.The trick is to export the effects
individually by right-clicking on them.
So you get a separate FX file for each effect, eg. Ground.fx,
GroundReflection.fx.
My 'groundReflection' effect
currently comes out black in the preview in RM.Nothing to worry about.
Started having thoughts about
adopting Max / RenderMonkey's 'Z is up' coordinate system.Decided not to, because D3D uses my 'Y is
up' system (eg. it converts meshes to this system).If I'm going to have to make conversions,
I'd rather be doing it in exporters than at runtime.Just a bugger for RM previewing really.
RM: Added an INGAME define
(undefined in RM, defined in game).
Started distoring the ground
reflection based on the surface normals (currently just using the ground's
vert normals).You can see it's
*trying* to work, but it's definitely not right.
TODO: try displaying the plane-space normals
(or the UV offsets) as RGB, to try to work out what's going on.
TODO: Jazz up the website a bit
when I upload build #3.Stuff like the
'total hours' count would be nice, even this whole devlog.
17-Jan-2006
2
TODO: Read up on VSS web project
stuff!!Looks cool.
Gave PPALMER VSS 'destroy'
permission.
Tried to set up my VSS
'website\final' folder as a web project.But when I try to deploy my site content, it just sits for ten
minutes, overwrites my online index page with an empty file, then complains
that it couldn't connect to the ftp site.The ftp site that it's just finished mauling.
15-Jan-2006
8.66
NOTE: RenderMonkey doesn't seem
to know about multi-element textures (ORTs with two layers of colour).
But my card seems happy enough to create such an ORT for me though
(D3DFMT_MULTI2_ARGB8).
IMPORTANT: In a PS, can't be
testing values of the current render target to decide whether or not to draw
the pixel!It causes a contention
problem and gives you pretty random results.
This was buggering the sub-ground masking for the ground reflection.Took me ages to work this out though.
My road lettering came out as
paint-texture blocks when I switched the patchwork texture to 64 bit.
Had a good old think about the
ground reflections and how best to proceed with them.
- Will continue to render from the point of view of the reflected
camera.I considered rendering from
the POV of a point on the reflection plane (making all the necessary mapping
adjustments), then realised that would be bad POV (eg. what if POV was right
next to a kerb - the kerb would block out buildings).
- Reflection render FOV should be flexible (up to 90° say).I'm willing to render a five-sided cube map
of the reflection geometry if need be.This would be oriented with the reflection camera.The minor sides would be much lower detail
(eg. no characters, or lights only).
- The final surface normal of each pixel will divert the mapping of the
reflection image.I will probably have
to scale the diversion based on the depth of the pixel, for perspective.This is sound for building / streetlight
reflections but not for the moon.
- The ground reflection plane, if I do decide to keep it (probably should)
needs to morph around very discreetly.
- I'll want to figure out a way of reflecting the kerbs, later on.It looks good already when you get what
looks like a kerb reflection.
Did lots of bits & bobs
(fixes) on the ground reflections.Pinning the reflections to the undulating ground was alarmingly
easy!Very similar to the patchwork
mapping :D
Having been playing around with
blue-sky, yellowish-black silhouette ground reflections, I'm feeling very
drawn towards adding a daylight mode to the game (probably non-conflict; see
wishlist log).Because it looks eerily
lifelike already.
I think it would be very little work and would be an awesome thing to
include (like - survey the carnage early the next morning).
So from now on I'll be trying to make sure the stuff I do allows for
daylight mode to be added later.
TODO: Get the character
reflections working again.
TODO: Get the road lettering
working again.
14-Jan-2006
12.33
NOTE: effects are not device
objects, but they do/can contain device objects.
Got my head round the
InvalidateDeviceObjects / RestoreDeviceObjects / DeleteDeviceObjects lark a
bit more.
I call all the restores at the end of the game initialisation, as if the
device had just been changed.
Decided there was no reason to
force device-dependent initialisation into its own stage.Nothing should ever initialise during an
absence of the device.
Found that you can (and should)
set effects' texture parameters rather than manually setting the texture you
want on the corresponding stage (which doesn't seem to work properly).
Need to NULL the texture parameter on losing the device though.
Had to start splitting my shader
effects into separate rfx files (ie. one file per effect).This is basically because
D3DXCreateEffectFromFile only supports one effect per file.Also it will cut down on variable name
conflicts in the fx file.
Managed to get the second pass
of an effect working.Just need to
make sure I set all the texture parameters that are needed.
TODO: Get that 'clip' working in
the GroundReflection PS.
13-Jan-2006
5
Got the true ground effect plane
working (currently just the player's ground normal, as a stand-in).Had to adjust the shaders for this.
Took quite a while to get the patchwork decal object matrix right, but I
got there in the end.
NOTE: There's a tiny decal culling bug - it's probably not taking object
matrix scale into account.
Fixed the RM-vs-game axis
difference by defining _Y and _Z macros (as z and y respectively) that the
exporter converts to y and z respectively.
TODO: Get that ground reflection
rendered onto an ORT - I need to see it working!!
12-Jan-2006
7.33
Removed CModePlay::updateFlags
because I couldn't see a good use for them.(Why did I add them?)
Moved all the environment stuff
from CModePlay into CModeEnvironment, having decided I wanted to allow the
possibility of rendering the environment as a background for the front-end
menu for example, or during the credits, or whatever.Fairly easy change.
This is also in preparation for multiple environment subset renders (eg.
reflections).
Accidentally got a sort of a
golden-shine-and-running-paint effect from the patchwork PS by offsetting a
second layer of colour according to the natural red component of each pixel.
CAUTION: Some of those D3DX
functions, eg. D3DXMatrixReflect, do unwanted normalisation etc.
Changed all my vector3 functions
to take vector pointers rather than references.This is the way the D3DXVec3 functions are
done.
Replaced SETVEC3 macro with Vector3Set function.
Started working on reflection
code.This is initially for the
ground, but I expect windows etc. will follow soon after.
TODO: Get the view set up
properly in CModeEnvironment::RenderReflectionImage
11-Jan-2006
3
Got rid of all the RM FX name
mangling problems by auto-remming-out variable redefinitions (see define
NOCTAMBULEFX_REMOVE_DUPLICATE_VARIABLE_DEFINITIONS).
From now on, can't re-use same name for different variables (all shader
variable names have to be unique).
I'm still mangling struct names etc.
Had to add a fix to stop
function names getting un-re-defined (exporter didn't know the diff between a
variable name and a function name; now checks for '( ' after the name.Function names are mangled.
Continued building up the ground
effect in RM, improving the exporter as necessary.
10-Jan-2006
5.25
Note: texture surface levels
have to be released when you're finished with them.
The name mangling (of the matrix
mentioned above) seemed to have arisen from the matrix being prototyped in
the VS (where it wasn't used) as well as in the PS.
Fixed a bug (" true" /
" false") in RenderMonkey's FX exporter; added a new exporter
"Noctambule FX Exporter".
Moved the RM plugins into "noctambule/vss/RenderMonkey
SDK".
Added the FX exporter code to sourcesafe.
Wasn't able to get .NET-SS integration working on the FX exporter project.
Noticed I don't have permissions
to permanently delete files in the Noctambule VSS database.
Changed the ground effect to use
the proper "ground effect plane" info.This bypasses the axis-difference problem
mentioned above (for now).Next time I
run into that problem I'll adjust the exporter to correct it.
Disabled RM FX exporter's shader
variable name mangling (see define NOCTAMBULEFX_MANGLE_SYMBOLS).
TODO: Resolve remaining
name-mangling problems (ie. problems arising from NOT mangling the names).
Started setting up the rest of
the ground effect in RM - adding more passes.
Some of these are dummy passes, so I've changed the exporter so it doesn't
export passes with "dummy" in their name.
TODO: Take a field trip to
France to celebrate 500 hours of work on the project (time flies!)<-done
(june 2007)
13-Dec-2005
5.5
I'm totally ECSTATIC to have got
the ground effect working at last - I was beginning to worry that it might
not be achievable.
Did this by moving the projection from view to clip space into the pixel
shader.
It's like parallax mapping but using a mesh instead of a height map.Looks gorgeous.Even the artefacts look gorgeous.I'm so proud :')
RenderMonkey was mangling the
name of one of the matrices, which was stopping the game version working for
a while.
TODO: Sort out the
rendermonkey-vs-game axis difference, and see if there's a way of preventing
the name mangling problem.
TODO: Tweak the placeholder
lighting and slap a new build on the site??
11-Dec-2005
8.5
Spent the whole weekend trying
to get the shaders right for mapping the patchwork on.
Landscape\Ground\p0 is the best I've come up with so far but it needs some
kind of clipping.
I need either some better way of preventing perspective-correct texture
interpolation, or some way of clipping pre-projected vertices with W's of 1.
10-Dec-2005
9
Set up a RenderMonkey
'Noctambule' workspace with the understanding that this will contain ALL my
shader effects.
IMPORTANT: Had to switch the
'matViewProjection' (and 'matView') matrix in RenderMonkey from
'viewProjection' semantic to 'viewProjectionTranspose'.
So it would appear that D3D and RenderMonkey use opposite matrix formats?!
NOTE: Can't view rendermonkey
.fx files in the D3D effect viewer, for various reasons - not a problem to me
since I'm happier using rendermonkey.It supports drag&drop for a start!
NOTE: RenderMonkey is a bit
flakey with its compile error messages.
NOTE: In rendermonkey, had to
change my textures' origin to be 'bottom left origin', and my models'
orientation to be 'RHS coordinate system'.
Wasn't able to run anything
other than pass 0 of my ground effect (had to move the pass I wanted to run
into pass 0)
NOTE: RenderMonkey seems to
share Max's weird Y-Z swap.Will need
to deal with this in some nice way.
TODO: zwrite pass for the ground
so it sorts with the buildings.
8-Dec-2005
5
Started reworking the patchwork
system.
Meshes now transform all their
verts into world space, in CalculateBounds.Saves having to set an object matrix when drawing each mesh.
Added special culling for ground
patchwork decals - projects the object onto the ground effect plane and culls
it from there.
TODO: optimise by taking advantage of the fact the object has no height.
Got the ground decals rendering
on their ground plane to the ORT.Need
the vertex shader next to map it onto the ground.
5-Dec-2005
1.833
TODO: Re-work the patchwork
system as follows:
- Render all visible decal objects onto the approximate ground plane, on a
screen-sized render target.
- Use a vertex shader to project each ground vert onto this plane, thence
into screen space to get the UV coords for using the ORT as the ground
texture.Camera-space texture matrix
might be of some use.
This will solve a bunch of problems, most importantly it'll give me the res
I'm currently lacking.
The approximate ground plane is the same one used for the ground
reflections.
Fixed the ESP problem by linking
with d3dx9d.lib rather than d3dx9dt.lib.Tom Gaulton suggested it could have been a bad mix of libs that was
causing it, and he was right.
Eventually got some sensible
results out of the 'BasicHLSL' FX - phew!
I'll now start cannibalising this into the new patchwork effect for the
ground.
As soon as that's working, I'll set up the ground reflection effect to pipe
it through :D
It'll be ACE!!
4-Dec-2005
9
Re-installed DX9 Aug 2005 - this
gave me back my shortcuts etc.
Installed RenderMonkey - looks
like a really good effect editor and has tons of example effects.
Clouds - see RenderMonkey
example "Reflections Refactions\OceanWater_ASM"
Got the missing DX code stuff
back in - must've been the uninstall attempt that removed them.
Started planning the custom
per-pixel ground lighting / reflection.The RenderMonkey "Reflections Refractions" example is a very
good reference, especially with the 'bumpiness' value turned up high.
Had to remout the blur offsets array from the final pass of the reflection
effect to get this workspace to export (?)
Started adding .RFX
(RenderMonkey) files to the project and SourceSafe.
Can't launch RFX files from
.NET.RenderMonkey opens but not with
the document.
Got rid of the 'unicode'
configs.The normal configs are
unicode now.
NOTE: Use
UtilCreateEffectFromFile instead of D3DXCreateEffectFromFile.
Couldn't get any effect to
work.Calling any function on an
effect is giving me "value of ESP not saved properly..." error.
3-Dec-2005
5.75
Found that some DX stuff
suddenly didn't exist (probably because of the attempted uninstall then
reinstall):
D3DXComputeTangentFrameEx;
D3DXFX_NOT_CLONEABLE.
Luckily the game seems to run fine without them.
Found that all objects in max
need to have their object colour set to full white, otherwise their object
colour always pollutes the exported vertex colours.
That is unless you use the 'ambient diffuse override' thingy in the
exporter.
Found that ampersands should not
be used for naming obejcts in max.
Found that after using max for a
while, the game can fail to create a d3d device.Closing max fixes this.
Cleared the node colour of
everything to white.Told max to use
white for all new nodes.
Got rid of placeholder solid
colour materials - shouldn't need this kind of thing again now that I've
sussed the colour lark out.
Added a 'PointLight' plugin for
dynamic point lights.
Added CLightManager, the class
that stores and manages all the dynamic lights.This is now parsing the PointLight
instances nicely.
Spiced up the website with some
keywords.Found that it is showing up
in Google now, even if you limit the search to English results or French
results :)
2-Dec-2005
2.5
MaxScript: Found that spinning
'objects' only gives you the non-hidden nodes.If interested in cameras, you should spin
'cameras'.
PhotoPoint: Added a UTIL button
for doing general fixup stuff.Used
this to correct the paths of all the photocam photos.
Although this worked, leam.max still contains those same
"\\newdell\p_drive " paths.No idea what that's all about, but it doesn't seem to be causing any
problems.
Decided to ditch the P: drive
and move the Noctambule folder into the root of C: instead.The P drive was a brilliant idea in theory,
but in practice it caused more problems than it solved.Those network folders just don't work as
well as they should.
Anyway, this changeover went frighteningly smoothly.
M & G confirmed that the
build (v2) downloaded from site runs perfectly :D
30-Nov-2005
2
Removed the redundant lib copies
from run.bat; checked that it still works.
I reckon the old lib copy should do no harm though, even on new DX
versions, since the libs are numbered according to version.
Found that DX9.0c runtime can't
be uninstalled - it even says so on the Microsoft site.So I wasn't able to install a French
runtime to test with.
Added the batch files to source
control.
TODO: I've asked IT to let me
change my computer name back temporarily.PhotoCams need to save the FILENAME of their bitmap (UPDATE - they
already did), and this needs to use "P:\" rather than "\\pc375\p_drive\".Test this by changing the computer name
again.
Existing photopoints in leam.max are completely broken till I can change my
machine name.
11-Nov-2005
6.75
Had to move all my folders
containing source-controlled stuff into P:\Noctambule\VSS, in order for .NET
SourceSafe integration to work.
I think it's because of IT mucking about with network stuff.They even changed the name of my PC so I
had to re-map drives etc!
Found that I could control the
game using Tony's RC aircraft controller (eventually worked out that this was
why the controls seemed jammed!)
Got the game running in the new
location (VSS folder).
NOTE: Files deleted from the P
drive (or any 'network' drive) get properly deleted, not sent to the recycle
bin!TODO: Find a way to change this?
Found that the .mdmp crash dump
files that get created for Microsoft are quite useful for debugging crashes
that are otherwise hard to get to (eg. fullscreen-only crashes).
Had to add d3dx9dt.lib to the
project's linker input for Master and Release configurations.It was complaining that the likes of
IID_IDirect3DBaseTexture9 were unresolved external symbols.
TEMP?: Disabled the F1 (control
config) key in full screen because it was causing a hang that I couldn't
really debug.
Found that the game clamps the
window size, so the default window size works fine on any res :)I'll therefore probably leave windowed as
the default until the game nears completion.
Re-arranged the file/folder
hierarchy of the install because the game wasn't finding files properly since
the shift to the VSS folder.TODO:
improve the file finding!
Put build #2 on the website;
updated the website.Added some more
folk to the mailing list.Submitted
the site to Google.
3-Nov-2005
Found a util for making font
textures out of font files: C:\Program Files\Microsoft Xbox
SDK\xbox\bin\fontmaker.exe
29-Sep-2005
3
Set up a memory-depletion
handling function, 'HandleMemoryDepletion'.
NOTE: The 'unreleased device
objects' error box is displayed from DXUTCleanup3DEnvironment (DXUT.cpp)
Remembered about the AXIS(vec,
axis) macro.
Set up a 'NEW' macro (to replace
'new') which causes leaks to display their origin in the source.Should nearly always use NEW instead of new
now.Defining 'new' as something caused
problems for classes trying to override their 'new' operator.
Fixed the unreleased device
objects - they were temp surface obejcts created by a
'DEVICE->GetRenderTarget' in CPatchwork::GenerateTextures.Now calling Release on the surface when
finished with it.
It was also a 'texture->GetSurfaceLeve' in
CPatchwork::ApplyTexture.Bizarrely,
this Release returns S_FALSE rather D3D_OK (??)
I can get into full-screen now,
but I can't get back out…
TODO: get that surface level
mentioned above, in the restoredeviceobejcts function rather than where it is
now.
Moved the landscape shutdown
into its destructor (from its shutdown method), because the shutdown method
was never getting called! (hence the leaks)
28-Sep-2005
5.5
Fixed the patchwork sorting
bug.So the patchwork system is up
& running now, that's good.
EXCEL: Ctrl-cursor to cursor
through text rather than between cells.
TODO: fix the misc. bugs in
order to get build 2 together; also set up placeholder lighting (the
beginnings of the real-time lighting system).
After that, probably pixel-shader-o'-clock :D
Added a default material,
'g_defaultMaterial'
TODO: Work out how the patchwork
is still leaking memory; get it alt-entering smoothly.
Got the zombies and crosshairs
drawing again.
26-Sep-2005
5.75
NOTE: The '* Unicode' project
confifurations are not to be used (the normal configs are now unicode
anyway).
TODO: get rid of the unicode configs.
Set up multi-band support for
patchworks.Road patchwork is
currently working pretty well with 3 bands.
Used the 'border' texture-addressing mode for each of the bands except the
outermost one (wrap).
Bunch of improvements to the
mesh classes.
Patchworks now gather up all the
decal frames inside the outermost band, then qsort them by height.Z write and Z test are now disabled when
generating the patchwork textures.
Inner bands then spin through that decal subset testing visibility and
drawing their visible ones.
TODO: Sort out the weird drawing problem resulting
from this qsort!
(D'oh! I was interpreting the qsort callback
parameters as mesh pointers rather than mesh ptr-ptrs.)
25-Sep-2005
5.5
Tried enabling 4-sample
multisampling for the window.Did what
it said on the tin, but also corrupted my ground patchwork and increased
tearing, so I disabled it again.
TODO: Only move the patchwork
ranges by multiples of their world texel size!This will completely prevent fizzing!!!!
:D
And it does!Yay!
EVENTUALLY sorted all the
problems with the patchwork texture matrix, but this took up most of my
evening.
Texture transforms are always a complete bitch.
Added multisample support for
patchwork texture generation, on an option (defaults on).TODO: check the multisampling is doing its
thing; check caps before enabling multisampling; do away with redundant
surfaces when the option is off.
TODO: Fix the lighting /
renderstate / flickering issues to do with patchworks.
I had been preventing materials (as well as textures) getting set when
drawing patchwork receivers.
This patchwork system is really
proving itself worth doing now :DWorks a treat.They'll never
know how it's done, tee-hee!
24-Sep-2005
10
Made placeholder pavement &
road textures, 'placeholderPavement' and 'placeholderRoad'.
Fixed the top-down orthographic
culling, phew!
Got the beginnings of the
patchwork system up & running.It
doesn't seem to like ORTs bigger than the primary render target, so it's
currently very fizzy.Longer intervals
between patchwork updates will reduce the fizz I suppose (currently every
frame).
I need to get the patchwork rendering looking decent or else I won't be
able to use it.
I'm worried about my patchwork height map / normal map fizzing - could look
really bad.
TODO: Get the patchwork system
to survive alt-enter; see how things look with a 1024*1024 ORT.
Investigate <device>::CreateRenderTarget, and multisampling for the
offscreen rendering (might help against the fizz).
NOTE: Looks like ORTs have to be
power-of-two dimensions
NOTE: 512 is a good res for a
4*4m patchwork range.I'm thinking I'm
going to need to do multiple passes of the patchwork render, to get a sort of
manual mip-mapping thing going on.Luckily, the patchwork system lets me keep the ground geometry nice
& simple, so multiple renders of it shouldn't be too bad.Maybe say 4m, 16m and 48m, all at 512.And of course, the bigger the square, the
less often it needs to regenerate its texture!Cool.There's hope for this thing yet...
Now defaulting to a full-screen
window - best of both worlds.
22-Sep-2005
2.5
Improved the camera system and
the view-info system.Well worth doing
- it's nice & clean now.
Added UtilTestXZBoundsVisibility, which is now used for CPatchwork's mesh
culling.*Nearly* ready for that
offscreen rendering…
21-Sep-2005
2.25
Investigated the official Max .X
exporter (C:\Documents and Settings\PPalmer\My Documents\Visual Studio
Projects\Max).Trouble is it's for Max
7.
I hacked the code up a bit and got it loading and running on Max 6, but
trying to export anything more complex than a sphere just crashes it.I couldn't see a way of debugging the crash
either.I could really do with a copy
of Max 7 - I'll look into that...
So I'm thinking I should try to
reconstruct the hierarchy myself based on bounds containment, or
something…
Actually no, a MaxScript plugin to write out a map of the hierarchy would
be better, then this can be reconstructed by the game.
17-Sep-2005
13
Did a first pass texture shoot
round Bedford Court, from about 7.20->8.20.Came out well - definitely stuff there that
will make it into the game.Timing
worked pretty well too; plenty of light but not too much activity.Might do more tomorrow
Cameras now have names; added a
CNameUser base class for this.
Got rid of old DebugCamera files
because they weren't being used.Added
a 'Source\Graveyard' folder for this.
Eventually sorted all my
hierarchical landscape bounding sphere culling problems.Took much longer than it should have
though.
I've been a bit sleepy from having got up so early for the texture
shoot.
UtilTestSphereVisibility is deadly accurate now.
Found that ImageAlign is a far
better tool for reshaping texture photos than Photoshop's (current) transform
tool is.
256*256, 8 bit PNG ~= 66K.That's good!
TODO: use normal mapping on
windows to give the impression of warped glass.
Had another play with the Flame
materials, to get me thinking.Stumbled across a smashing parallax refractive rain-bounce effect for
the ground!(parallaxRainTest.fmt).Uses a glass-smash texture that Steve T
did, but I'll want to animate or proceduralise my height map.
Added a 'Mockups' folder with screenshots of this stuff.
16-Sep-2005
2.75
Put atemp welcome page on the website.
Got a Zbrush crack from G.
I can’t stop admiring my new
baby website!It's so cuuute!
Jettisoned the DXUTMesh files,
copying the code instead into the Mesh files for extension.
Landscape: replaced a wad of
code to determine the mesh's vertex size, by using
D3DXGetFVFVertexSize(pCollisionXMesh->GetFVF())
Mesh: used
D3DXComputeBoundingSphere to get frames' bounding spheres and centres, but
they look like they're in hierarchical space or something.
NOTE: CMeshFile inherits from CMeshFrame!!!Duh!
Tried making sure the view
matrix _44 was set to 1 in case that was what was causing the ED view angle
glitch; no luck.
Unremmed-out
UtilTestSphereVisibility.
15-Sep-2005
3.5
Bought programmerart.org for two years, plus one year's hosting. :DNow I am the master.The web master.
DXUT.cpp: bodged-out a check
that was failing saying that I was running a different version of DirectX
from the one that DXUT was compiled on, or something (?)D3DXCheckVersion.
Got hierarchical mesh loading
and rendering working YAY!The
DXUTMesh file was confusing; turns out the CDXUTMeshFile class was the
hierarchical one, the one I should've been using.CDXUTMesh is a flat thing, used for leaf
meshes in the CDXUTMeshFile's hierarchy.
Been changing WCHAR stuff in the
UT code to be chars.
Photoshop's
'filter\texture\grain\clumped' gives a nice tarmac effect.'Wave' and 'Zigzag' filters can be used to
make lettering look hand-painted (TODO: apply this to the pavement font
texture).
CModeStack: Fixed bug with
Pop(); added destructor that empties the stack (I had some memory leak
warnings without this).
Jettisoned some 'indexed
shaders' stuff from CAnimMesh; old legacy junk from the MultiAnimation sample
by the looks of it.It was causing a
crash when I alt-entered, because it was trying to do Invalidate/Restore type
stuff with it although it hadn't been set up.
Made landscape collision be
generated only from the "ground" object.
Got hierarchical culling working
fine using object centres.
TODO: Get all this landscape culling working nicely (bounding sphere + XZ
bounds for top view) so I can get going properly on the patchwork system.
14-Sep-2005
4.5
TODO: Texture photography this
weekend!
If I'm going to be experimenting with the patchwork system, I might as well
work on textures I can keep rather than have to throw away.
Get tilable road/pavement surfaces, caps and covers, patches, seams,
cracks, etc.
- Clean pavement tile(s)
- Clean road tile(s)
- Big cover outside Bedford Court
- Bedford brick
- John St. sign?
Had to add d3dx9.lib to the
project's linker input.
Installed the August 2005 SDK
update, because I got a DXUT init message saying I was running "the
wrong version" (D3DXCheckVersion).
EVENTUALLY managed to get the
game running again, using CDXUTMesh for the landscape, and having jettisoned
all my old util files from the previous SDK.What a big old ball-ache it was though!
Decided it's easier to do all my
filename strings as good old-fashioned char arrays, and convert them to wide
strings at the last possible moment when necessary; added a nice lazy util
function for this - UtilGetTempWideString.
Been thinking a lot about a
procedural texture system kinda like the patchwork system, mostly for wall
text.
Would be awesome to be able to zoom in on kebab-shop menus and read big
car-park signs and stuff!
Tried to use
CDXUTMeshFile::CreateFromResource, couldn't do it.
13-Sep-2005
0
In .NET, ctrl-alt-P brings up a
nice 'processes' window that lets you attach to any process currently
running!
12-Sep-2005
4.5
TODO: remember to investigate
the 'shader debugger'.
Placed some more road lettering,
using Max's 'reference' copies.DISABLED & LOADING ONLY on Bedford Street.Very quick to do :)
Road markings make a lovely relaxing 5/10 minute break during tough coding.
Tried to use the CDXUTMesh
class.Found that the DXUT stuff
needed a unicode setup.Got into a
world of hurt trying to set this up - jettisoned all my existing DX util type
files from the previous SDK, and started setting the app up in the manner of
the new samples (looks quite a bit cleaner).More still to do...
The DXUT stuff looks good.I just
wish they'd make their mind up about this stuff!
TODO: Get that pixel motion blur
in for the next build.
11-Sep-2005
10
I'm now using self-extracting
zips for the releases.This is because
the temp storage folder on X:\ was deleting some of my files, when obviously
it's better for it to delete either everything or nothing.
Installed WinZip because
ZipCentral is a bag of wank.
Found a site all about roads and
road markings: cbrd.co.uk.This
contains downloads and info for all the road marking fonts :)
Got the 'pavement' font plotted
out using Xara.Just need to add
variation to the bitmap now, including alpha channel variation for
thresholding.
Added a 'pavementLetters' object
that I'll use to create road lettering.
Started setting up a
hierarchical mesh system.Eventually
stumbled into the CDXUTMesh class which appears to do exactly what I'm after
(even optimises the mesh).
TODO: Assimilate CDXUTMesh into CMesh!!!!!
Installed the DX file formats
plugin for Photoshop / PSP.Lets you
save fancy .dds files with mips and stuff.
10-Sep-2005
8
All file paths are now specified
relative to the game root folder.For example:
g_gameRootPath = "..\\..\\"
textureName = "Graphics\\Final\mytexture.png"
LoadTexture("%s%s", g_gameRootPath, textureName);
The game root folder is now
designated by the existence of a file whose name is defined by
GAME_ROOT_MARKER_FILE (currently "GameRoot.dat").
TODO: Stop using
DXUtil_FindMediaFileCch!!
Couldn't stop the background
plane photos exporting with leam.x.Even tried hiding every layer, but they still exported :(
For some reason, the giant quads
on the ground obejct were causing the HOP tests to fail (the pre-collide test
was bailing out; forgot to try skipping the pre-collide).Worked round the problem by putting a few
arbitrary cuts in the giant quads.
Found you can batch-build
projects by right-clicking on the solution item in the solution explorer.
Ran into a problem with the
shader data in the project resources.Got round it by removing the data and using D3DXAssembleShaderFromFile
instead of D3DXAssembleShaderFromResource.
Removed redundant options from
the app's menu.
Got V1 running on PC228.First it complained that it couldn't find
d3dx9_26.dll (this file didn't seem to be on the machine anywhere, despite me
having just installed DX9.0c.No idea
what that's about.So I made a
'run.bat' that copies this dll into the PC's windows/system32 folder, then
runs the exe.
PC133 only had dx8.1, so it
complained about a missing d3d9.dll I think.I couldn't download DX9.0c because Windows validation failed (invalid
product key).
Installed DX9.0c on PC142, but
then it restarted itself and I don't know its password.TODO: try it again on this machine.
Made a batch file to assemble a
build of the game, and a 'builds' folder to keep them in (recent ones at
least).
Made a release log (txt) which
gets copied into the build.This will
list the new features and fixes of each build.
Nervously released build #1 to
my small distribution list (in the 'contacts' category).
TODO: I'm now going to get
started on the patchwork system - YAY! :DThis is going to be a big, big, task though.
TODO: Try swinging the viewpoint
around according to the 'distanceToScreen' option?
9-Sep-2005
5
Had a good old think about that
patchwork idea I always imagined, for the ground texturing.
I'm going to go ahead with it, because it looks like
it'll be absolutely brilliant!
This system simultaneously solves all the problems of road markings, repair
patches, designed surface variation, manhole covers, muck, puddles, chewing
gum, fag-ends, tyre marks, etc., etc.It'll be ultra-powerful and should actually save frame time.
Restrictions: Probably won't be suitable for drains; won't work on
kerbs/walls.Only for streets and
pavements.A dummy texture (the
aerial) will be planar mapped in Max onto the streets and pavements.While the game's running, the decal layers
will be rendered offscreen and applied to the ground using a texture matrix
to shift the rendered area into position.I'm thinking there'll be one decal stack for the streets and another
one for the pavements - this is more flexible and facilitates kerb position
tweaks.
NOTE:Before loading Max, you've
got to disable all E-D's D3D support, otherwise the viewports get corrupted
unusably.
Found that, when exporting the X
file, it's better to switch off texture conversion and let it use the
original textures.
Considered GetMapping's digital
surface model again.The thing that
makes it useless is the 5m horizontal posting - all the streets would slope
up to the roofs like a snowdrift.
Got the placeholder crosshair
drawing, using an orthographic projection matrix.The thing I'd forgotten was to make the
VIEW matrix identity as well as setting up the projection matrix.
Looked at the 'Motion Blur'
sample - it's gorgeous!
TODO: PIXEL MOTION BLUR!!!!!
Fixed a Release/Master FOV bug
(wasn't being updated each frame and currently it needs to).
Finally got the exe to run
without the debugger, by launching it from a shortcut whose 'start in'
directory is set to 'P:\Noctambule\source'.
TODO: Top priority is to get
that portable master build working, then maybe have fun with motion blur or
the patchwork system.
TODO: Placeholder gunshot sample
(get the audio system started).
Note to self: Noctambule.org is
not work-safe! :/
8-Sep-2005
5.3
Tried all the deformation
modifiers that looked like they might help with my ground-detail vs
ground-slope question.
SurfDeform WS came closest, but it would've given me the same problems that
the meshsmooth did - not enough control basically.
Luckily it looks like soft selection will do what I need anyway.
Found the E-D fly-cam bug: the
fly-cam was setting up view & projection in PreRender, then the debug
mode was drawing text in Render, disrupting the subsequent play mode
Render.
I've changed the mode stack to render from the bottom up, which is always
the way I thought it should work anyway.
TODO: start rendering from the lowest opaquely-rendering mode, if any.
Added a 'Master' configuration,
which defines the '_MASTER' symbol.Master config will be used for all the proper builds that I
distribute.
The 'Release' configuration now
defines the '_RELEASE' symbol.Release
config is optimised like a master, but still has timers and debug
output.It's essentially for assessing
the performance that the master builds will have.
The debug symbol to use is
'_DEBUG'.
Added CModeHUD, for crosshairs,
gauges, etc.Tried to get a
placeholder crosshair drawing; couldn't get the object matrix right, but I'll
try again.
TODO: GET THIS WORKING, AND RESTORE PREVIOUS OBJECT MATRIX FOR THE DEBUG
RENDER!!!!
TODO: Get a presentable master
build together asap, to prove that the project exists!
1-Sep-2005
3
Borrowed some LCD-shutter 3D
glasses and tried them out on the game - looks great.
Currently lots of ghosting though because of the gaudiness of the
placeholder buildings.
This glasses system doesn't let me do any of the stereo myself - it does it
all its own way :/
Scroll-Lock toggles glasses commands; F5 = on, F4 = off.
Improved the options
system.FOV is now calculated from
physical measurements in the options class so that it matches with the
player's real-world FOV.
TODO: Fix these shutter-glasses
bugs:
When using the glasses: Landscape disappears in fly-cam mode;
Sometimes get an angle glitch when turning the view while running.
Checked that the 16:9 seems to
work.I'd love to try it out on a big
widescreen TV some time though!
31-Aug-2005
2.5
Got my first genuine in-game
fright while standing outside Bedford Court eating my dinner and admiring the
car park wall.Glanced right, straight
into the blank face of a placeholder zombie - completely forgot what it was
and why it was coming towards me!
Tried to map an N: drive to the
Noctambule folder - didn't work.The
folder didn't show up as an available mapping location for some reason (???)
Found an interesting DX help
topic: "Introduction to the 10-Foot Experience for Windows Game
Developers".Have a look at this
later and see what can be added for the 'Media Center' (sic) setups.There's a download for the relevant
extension SDK.
Added Options.cpp&h, which
include likes of stereoscopy and widescreen settings.
Got the basic anaglyph
stereoscopy working, using the D3DRS_COLORWRITEENABLE renderstate.Didn't hit the displayed fullscreen
framerate, so I'm guessing that's clamped (PHEW!).Anim system will need to be made
stereo-friendly.
Bedford area mesh tweaks.
28-Aug-2005
8
Decided to ditch the mesh-smooth
modifier on the ground.It was too
vague; didn't give me enough control, etc.
Mesh-smooth is removed after version 130 of Leam.max.
TODO: Remember to set up local
space (pivot) for each building.This
makes it much easier to flatten walls, etc.
Knocked-up a placeholder Bedford
brick texture and splatted it onto some objects just for fun really.Looks kinda cool.
TODO: When photographing
textures, it'll be a good idea to use some kind of scale marker
TODO: Realised it'd be far more
sensible to export collision geometry using MaxScript rather than fishing it
out of vertex buffers.
For a start, the collision will eventually be a separate (invisible) mesh;
secondly the plug-in could exclude polys inside collision boxes, etc.
UVW mapping modifier has been
really weird - seems to come out differently scaled on different objects
(?)
Also seems to get twisted up somehow after the verts have been edited.
NOTE: Hidden objects don't
export.
Started building details into
Bedford Court, as sub-objects.Should
probably shoot some detail reference and textures for this area soon.
Updated the 'world' file for the
new town.
TODO: Find out how to display
hidden verts!
27-Aug-2005
7.5
Noticed that the edges of the
ground had shifted out of position a bit - dunno how, but it didn't cause a
problem because of the planar mapping modifier.
Got ground collision working
nicely, for the player and the zombies! :D
Currently colliding with every poly in the world, but I'll fix that later.
26-Aug-2005
2.5
NOTE: Must avoid exclamation
marks in node names (x parsing doesn't like it)
Got the game running under the
new SDK.Had to remout my DgbAssert
calls, and use a different assert function, and remove an old library
reference from the solution settings (strmbasd.lib).
Got the new Leam model into the
game - this was very good for morale, and makes it much easier to judge
ground shape, etc.
Worked first time.Used the latest
Panda .X exporter.
TODO: Ground collision, even if
only placeholder.I want to be able to
walk around the new town as soon as possible, and I want the zombies to do
the same.
(frame 'ground', child currently "Object #16')
See ID3DXMesh::LockVertexBuffer / LockIndexBuffer.
Added a 'breaches' log to the
Schedule, which should help reduce absences from the project, or at least get
a clearer idea how they arise.
7-Jul-2005
6.5
TODO: keep the graphical
time-stamp on photopoint photos - it's handy.
Shift-click a photo thumbnail to
view the bitmap.
Built a bunch more bits into the
Bedford Court area.Found that it's a
good idea to stick lampposts etc. in to help with the photopoints.
Jogo showed me some stuff like
the target weld and using animated cameras to look around, in the absence of
a proper fly cam.
He says that Andy Slater wrote a Max fly-cam but it was slow, etc. - I
might ask about that.
Got a bit handier with cuts,
slices and booleans.
7-Jul-2005
3
Had a look at the samples on the
very latest DirectX, with my new graphics card.The post-process sample is especially cool
- this should show me how to do the colour-draining I need for the
stereoscopy.
Had a look at panoramic tripod
heads at www.kaidan.com - too expensive.
Did more Bedford Court
area.The plan is to build it into an
early prototype area - I want to see it in the game as soon as possible.
Had some problem auto-fovving
pp19 - I seem to remember zooming out & back in again during shooting
that one (which can alter the fov).Maybe that's all it is?
TODO: re-shoot, delete photopoint and strart it totally fresh - see if
there's still a problem.
I've got to start adopting a
more happy-go-lucky approach to the blocking-out process, since things still
aren't matching properly.To be
honest, they don't really need to, it's just me being pedantic as usual.
Anyway, I blame the aerial!It'll be
interesting to see what the re-shot aerial looks like; it's due this
year.Who knows, might be higher res
and better to use.
5-Jul-2005
2.75
Continued in the Bedford Court
area - going pretty well.
Downloaded the latest DirectX.
3-Jul-2005
2.5
Camera tilt now takes the
photopoint's full orientation into account.
Tweaked the ground and buildings
a bit round Bedford Court.
28-Jun-2005
4
Found what seems to be another
aerial splicing problem at the bedford/regent junction.Really can't seem to get that Jessop's shop
corner into place and agreeing with the aerial.
I'm now taking the aerial with a big pinch of salt.It seems that the best way for me to work
is to use photopoints to check the positions of other photopoints, forming
trustworthy clusters like the one at Bedford Court.
Cross-referenced the various
photopoints at Bedford Court, to improve the buildings and add more detail
(eg car park wall).This area is
really starting to fall into place, so I'm tempted to focus in on it a bit.It's very good for morale to actually get
stuff built and add a bit more detail to an area.Adding more of the buidlings round there
will help me to judge the ground shape too, which in turn will help me to
position more photopoints.
Used the new auto-fov-calculator
thing again - works an absolute treat!
TODO!!!!!!!!!!!!!!!!!!!: Camera
tilt needs to take photopoint's full orientation into
account!!!!!!!!!!!!!!!!!!
26-Jun-2005
8.5
Re-shot all my photopoints as
continuous horizontal tripod pans, between about 5.15 and 7am.
Used new pp17 and pp15 to tweak
the fov again to 49.306962703711542917626673274527 - sorry!
Came to the conclusion that
there is genuine (real-world) variation in FOV from one photopoint to the
next (due to slight inaccuracy in the NIkon's zoom mechanism).
So I abandoned the common FOV.When
welding a photopoint, you can now state its total angle span at current FOV-
the code then calculates the photopoint's own fov, and re-angles all the
cameras accordingly so that the gap/overlap is sealed.
This took fucking ages to get working - had to try every possible way of
rotating those cameras :(
This was tested/first used on pp12.
Photopoint must have ZERO ORIENTATION for this to work.
TODO!!!!!!!!!!!!!!!!!!!: Camera
tilt needs to take photopoint's full orientation into
account!!!!!!!!!!!!!!!!!!
25-Jun-2005
7.5
Measured the height of the
camera on the tripod, legs fully extended, pole fully extended, as
154cm.This is the way I'll be
shooting my photopoints from now on.
Used my most precise pan so far
("ppCalibration") to re-measure my photopoint FOV as
49.494671571448866999761693707093.
I will NOT be re-measuring the FOV at any point in the future - it was only
two hundredths of a degree out!!
I'll be using 8-shot tripod
panoramas from now on.I've added a
mark to the tripod so I can get zero pitch, and the tripod already has the 8
yaw marks that I need.
Added a global camera roll
setting to photopoints (one per point).Works well.Tried making it a
static member of the plugin but for some reason it didn't want to carry
through from each session to the next.It might be better as a global thing for all photopoints...possibly.
Made some photopoint fixes.
I'm going to hold off on the
common pitch angle parameter for now - I'll see how it goes just doing
horizontal pans.
18-May-2005
5.5
Worked out that only one
photopoint can allow free lookaround at a time using the photo planes.One other will be available for static
display of a single camera using the viewport background method.It will use the far-clip setting to cull
the other photopoint's background planes.
Re-worked the photopoint code so
they display a 'pan camera' for looking round the panorama (now up to 10
planes), as well as displaying the current photocam.This is really groovy.Mapped FOV to the D key for easy zooming.
TODO: Display previous
photopoint's current camera using the viewport background method.When a photopoint is deselected, maybe the
closest photocam to the pancam should be made current.
TODO: keep photocams arranged by
angle.
TODO: I'm going to re-shoot all
my photopoints as continuous panoramas, concentrating on the bedford court
area.It's such a superior method that
it seems silly not to.
I'll use the tripod for these.It's
a shame my tripod rotates its pitch angle by its yaw angle rather than the
other way round.I'll have a look for
one that works the other way.
Actually no, come to think of it all I need is a common pitch angle
property (relative to the photopoint).By tweaking this I should see all the photos twisting to meet each
other.Also, a common pitch angle
property would be a good way of counteracting any tripod-top tilt, rather
than doing this on the bitmaps.
Added a 'show buildings'
checkbox to the photopoint rollout, but it doesn't seem very responsive
(hiding/showing the buildings layer).
TODO: Add the common rotation
properties above, and test them in the warmth of the office.
17-May-2005
5.5
PhotoPoints now get forced into
the correct group whenever you select a photo.They were unhappy when I tried to do this
on creation or rollout open.
TODO: photocams should keep a
record of their native orientation.You should be able to zoom and pan them, and they should stay like
that till you tell them to 'return' (button).Need a 'Set native orientation' check-button that defaults true.The state of this should be saved with the
camera.
TODO: Global hide/show
ground/buildings check-buttons on the photopoint rollout.Keyboard shortcuts for this would be great.
Quite tempted by the idea of
buying some web space and a domain name.
It would be cool to put latest builds on there as well as having this
devlog online.
programmerart.co.uk and programmer-art.co.uk are both free,
as is noctambule.co.uk.
I could even use it as a data backup area if there was enough space,
hmmm...
Found that overlapping photocam
photos are useful because you can create a continuous panorama.
Used a photopoint panorama of
the office to measure a horizontal FOV error of -0.0146484375.Doesn't sound like much, but it causes
quite a big gap.Used Max's panorama
exporter to measure the gap in pixels (45 / 3072).
Accordingly, I've now adjusted the photopoint FOV to
49.516357264618434093161546085233.
I've also adjusted the photopoint background planes accordingly:
length: 489.15462886217099641210934067868
width: 652.20617181622799521614578757157
distance still sqrt(500m)
Leam.max version 83 is the last version with the old FOV.
I'm going to use the tripod for
all photopoint photography from now on.That guarantees that there's no difference in local Max-Y angle, and
if desired, no difference in local X angle between the photos.It also makes the photo height more
reliable.
I'm going to use panoramic
photopoints wherever possible - this allows me to ascertain the relative
directions of the cameras first so I can then just rotate them all as a
group, and also rotate the photopoint (to recreate ground shape) if
necessary.
Added show ground / show other
photos options for photopoints - helps when lining things up.
TODO: BUG: Photocams currently
don't transfer their roll angle to the background plane at the point when
they're viewed.
Got a good feeling about the new
field of view - things seem to be fitting better.Maybe it's my imagination.
TODO: Marks on the tripod to
indicate the rotation steps to use.
TODO: have as many background
planes as there are photocams, so you can pan round the whole panorama.
16-May-2005
5
Started getting scary messages
from Max saying that the 'm_photoFilename' property of the photocams didn't
exist.Turned out to be just a script
bug - phew!
NOTE: If Max's MacroRecorder is
enabled, with the 'Explicit sub-object sets' option ticked, moving an edge
will cause a crash.
This heppens even if the listener is closed.
Got the photopoints working with
textured planes rather than viewport backgrounds.
Now forcing photocams into the
photocams layer.I get an error when I
try to force photopoints into the photopoints layer though - will fix that
next time.
Layers rock!I can turn the entire building layer into
wireframe/see-through and still see the textured ground layer / background
photo plane!
They also let me hide away all those photocams which tidies & speeds
things up a lot.
TODO: Ability to view several
photopoints at once - ideal for cross-referencing 'live' while tweaking
buildings & slopes.
Made a tiny fix to the cropping
in the 'cropDebarrelled' photoshop action.
15-May-2005
6.25
Put the mesh-smooth modifier on
the ground mesh, and this (touch wood) seems to fix the HOP test problem.
NOTE: It's great when you can
see one photopoint from another - try to arrange this more often.
Deleted some older photopoints
with unreliable footpoints.
Got a black screen crash in Max
- most likely video drivers again.Got
to remember to save save save.
NOTE: Max: CTRL-S doesn't work
while panning a camera view!
Added the rest of the
photopoints I shot yesterday morning.There's currently 30 of them, taking up about 40 meg - won't break the
bank at all! :)
Should probably use the
6am->7am period for shooting detailed stuff; some of my current
photopoints clearly didn't have enough light and are blurred.Will also use the tripod & timer for
texture photography.
Couldn't see a way of setting
different backgrounds for different viewports via maxscript.Might well switch to the textured-plane
idea.
14-May-2005
10
Took a load of pre-planned
photopoint photos, between 5 and 6 am when the streets are empty.Could possibly have done them an hour or so
later when it's brighter - maybe try that next time.I'm guessing people start earlier on week
days than at the weekend though.
NOTE: Can't do conditions in
plugin rollout definitions
Made a whole bunch of
improvements to the PhotoPoint code, including handling for missing files.
Now grouping photopoint photos
by index; index can be set on the rollout.
NOTE: MaxScript editor's undo
only holds one item - so if you make two mistakes in a row you can't get back
to where you were! :(
Upped the de-barrel setting to
27 and updated the photoshop action accordingly.This is a 50% halfway-house between
calibration.png and DSCN1745.JPG - works well for both of them.I reckon I'm probably getting pedantic
about this de-barrelling lark...
I'm not going to worry about the out-of-date debarrelling on the existing
photopoints at all [yet].
Set up a load of new
photopoints.Had to re-jig the
existing ones because of changes to the camera class.
Still getting that
misread-ground-height problem.Looks
like it's just caused by testing giant polys, so I'll start dividing the
ground up a bit more.
13-May-2005
7.75
NOTE: Perpendicular photopoint
cross-references are the best kind.
TODO: Might want to phase-out
any photopoints that don't have clearly-visible-on-the-aerial foot
points.
Certainly need to pick my footpoints on the aerial first, rather than
taking the photos then trying to find the footpoint (which might be obscured
by a car or a shadow etc.)
Impoved building placement;
added new buildings, added more slope detail.The photopoints are really starting to make sense - it's all going
well :)
Found that Max can fail testing
the ground height after the ground mesh has been edited.Added warning dialogues for this.
Restarting Max fixed the problem.
Found that Max can also mis-read
the ground height.Don't know how to
fix this yet.
Ctrl-clicking a photo thumbnail
now also selects the camera object.
Found that User View is a much
nicer way of getting around than Perspective View, mostly because the zoom is
predictable.
Pre-plotted a load of
photopoints in Max, then grabbed screen images of them into the camera so I
can remember where to stand.
This is definitely the way to do it.
As soon as the sun rises, I'll go and take the snaps - should be nice and
empty at the crack of dawn.
12-May-2005
4.5
Got hold of 'ImageAlign PRO' and
set up a de-barrelling profile (de-barrel set to 25.50, re-scale set to
-1.00).
Complemented this with a PhotoShop action 'cropDebarrelled' which fixes the
aspect ratio back to what it should be.
NOTE: ImageAlign previews are to be taken with a
pinch of salt.
Set up all the photopoints I had
photos for, and I'm barrel-correcting all the photos now.
TODO: Nice calibration grid
photo to make totally sure the barrel correction is ok?
Noticed on my way home that
quite a few building edges (in real life) aren't vertical or aren't
straight.'Venue' for example is bent
just the way it looks on my photopoint (post-correction), which is nice.
NOTE: The aerial photo is not
gospel!It's especially dodgy at the
points where they've joined two photos, eg at the Yorkshire bank.
I expect I'll do some surgery on it later…
11-May-2005
5.5
Got the photopoints aligning
with the ground - far too easy!
TODO: Later on, I'm going to
want to view angles from several photopoints at once.Perhaps ideally it'd auto-show all the
cameras that can see the selection being edited.
Placed more PhotoPoints (around
Windsor, John, Bedford) while tweaking buildings.
Made a bunch of improvements to
the PhotoPoint plugin.
TODO: Lock photopoints'
orientations - a number of times I got myself in a twist by rotating the
photopoint when I meant to rotate the photocam.
On the subject of
photopoints:
- Watch out for dips in walls (ie. non-flat walls) - they can throw the
matching off.Eg. west wall of
Blitz.
- Don't spend ages matching a single photopoint perfectly - cross-reference
with another photopoint and get a happy medium.
NEXT: Put a few more photopoints
in, then continue adding slopes.
Try out the barrel-correction plugin for photoshop - might help deal with
the fisheye problem.
10-May-2005
3.5
Got the new PhotoPoints up &
running - sooo goood!
Started replacing the old photopoints with nice new multi-angle ones.They're matching very nicely (especially
the ground detail) thanks to the new FOV.
Looked at the Megon
smash-up-a-museum physics demo.Nice
shattering physics glass.
Had a couple of Max crashes,
including a blue screen when trying to resize a photopoint viewport.This looked like a graphics card driver
bug, and I do seem to remember people saying Radeon drivers can be iffy.
NEXT: Need to get photopoints
auto-aligning to the ground mesh (probably on selection of the
photopoint).
See 'How To ... Move Objects to a Surface' in the MaxScript reference.
28-Apr-2005
4
Found that when you copy photos
around, their creation date changes :(
Maybe I'll be wanting some code to name them according to their creation
date, straight after they've been saved.
As a temporary measure, I've switched to using the modified date, because
they initially match the date taken.
Made progress on the PhotoPoint
helper.Got it creating the child
cameras.It can't store these as
node/nodeTab parameters though, because that would create illegal cyclic dependencies.
Had to change a bunch of paths
etc. to get stuff working on this temp PC - worth doing though because it's
such a lovely snappy machine.
Steve T gave me a quick
demonstration of Zbrush.Looks like
it's going to be an essential piece of kit for when I start texturing and
creating materials.
27-Apr-2005
2
Set up the project on my
borrowed ninja PC, which has a "Radeon X850 XT Platinum Edition".
Had a play with Flame's SM2
materials - very very nice.I can't
wait to get my hands on that stuff.
Codenamed the project
"Noctambule"; renamed some stuff accordingly from 'Zombies', to
prevent unwanted interest.
Also changed everything to live on a 'P:' drive (the ninja's D: drive
wasn't available).
Current priority is to get this
new photopoint helper working - then I can really get going with the
construction.
25-Apr-2005
1
Fiddled with the new photopoint
helper, then decided to go out and get pissed and see some bands - sorry!
23-Apr-2005
2.5
Measured the camera's field of
view, using blu-tac, embroidery thread and my new tripod (48.79102º x
37.57088º).
This is equivalent to a 39.689mm lens.Updated the current PhotoPoint class to force this FOV.
Some PhotoPoints are definitely matching better as a result (the House of
Fraser / Blitz one is pretty exemplary), but Windsor St. and Murphy's are
still weird and I don't know why.I
might well end up having to distort the aerial on XZ, but I'll probably
consider it a last resort.
Measured the height of the
camera on the tripod (legs fully retracted, pole at full height) as 92cm).
Noted that the tripod currently
tilts the camera clockwise on Z by about 1.5º.
Noticed a fair bit of fish-eye
on my calibration photo - slightly worried about that...
14-Apr-2005
2
Started the new PhotoPoint
plugin (extends the Point helper to look after a group of photo cameras).
TODO: clicking on a camera photo
thumbnail should toggle that camera.Oldest camera should be replaced when you run out of viewports.Use a right-click menu for photo options (view,
browse, clear, etc).
11-Apr-2005
4
Really getting the impression
that my FOV is wrong.Going to measure
this empirically ASAP - totally pointless trying to match any more
photopoints till I have.
Started getting the PhotoPoint
plugin working (complete with photo & footpoint-photo thumbnail/pickers
:)
Needs quite a bit of work still, but it's going to be ace.This plugin lark is much simpler than I
imagined it!
10-Apr-2005
11
NOTE: See LODTester.ms for
examples of gui stuff.
Selecting PhotoPoints now does
the following:
- Changes the bottom-left viewport's view to be that of the selected
PhotoPoint.
- Changes the bottom-left viewport's background image to be the
PhotoPoint's photo (specified in the 'photo' user property).
- Applies the photo file's creation date&time to the sunlight system
(this was really quite complicated!).
Used a startup script for this.Lovely swish result.
Stuck a few more photopoints in
- got a bunch more ready to place.
Started trying to write the
photopoints as scripted plugins - this definitely seems like the way to do
things.Should try to get this working
next time.
Slightly less pessimistic about
the photopoints now.It's all about
cross-referencing - that's the only way it'll make sense.
Matching up the ground seems to be a fairly good way to start.
Just need to add lots of photopoints - got to get the ground shape worked
out, then things should start to fall into place :)
Learned tons about MaxScript
tonight - it's very groovy.
9-Apr-2005
13
NOTE: Camera batteries don't
hold their charge for long periods of time (eg. a fortnight).They need to be charged right before
they're used.
Decided on the world alignment
that I'll be stuck with throughout the entire project: Z -6.15°.It was a bit scary making this decision,
but I reckon it's aligned well so I'm going to forget about it now, take it
as gospel, and get cracking with the construction work.
Leamington Central Point is
right outside GameStation, 83 The Parade, CV32 4AY
Spent ages trying to match a
photopoint of Windsor St.Only had a
single photo because the cold killed my camera batteries.
Learned a few lessons from it
all:
- Before shooting, check that the from point is clear on the aerial (ie.
carry a printout).Arrow tip/tails,
give-way tips and centre line tips on the road seem to work best.Kerbs are not clear enough on the
aerial.
- Precede the shot(s) with a photo aimed exactly at the from point.
- Shoot the street from various positions and angles so you can
cross-reference.
- Try to get as many reference points / vertical lines as possible into the
shot.
8-Apr-2005
4
Made the world exporter loads
nicer and safer.
- now gives you a save dialogue.
- will never lock files when it hits an error.
- exports SpawnPoints.
- supports scaled rectangles (CollisionBoxes), which are much nicer to
edit.
- Added a .zwf format for it (just text still)
Learned some Max stuff.
- selection filter
- selection locking (space)
- handy manipulation shortcuts (q,w,e,r)
- Macro recorder
Added SpawnPoints.Currently zombies respawn at a
randomly-picked SpawnPoint, but that will obviously improve.
It's beginning to play now that
the zombies emerge from the alleyways.
TODO: Auto-expand collision
boxes - they'll be placed down tightly around buildings.
Fixed some shutdown leaks.
7-Apr-2005
3
Player collision with AA
collision boxes is working.
Added very rough zombie
once-per-frame routefinding, but it's too rough - you can get them stuck just
by hiding on the opposite side of a building.Need to write decent routefinding before it'll play.
Looked at Max's
sunlight/daylight simulator - should in theory allow me to match sun
positions and shadows with those on the PhotoPoints.Just need to remember to keep the camera's
time&date set properly.Might be
able to write a MaxScript to set the sunlight's time&date whenever you
select/preview a PhotoPoint?
Oh yeah, pretty much worked out
how I'm going to gather ground height data.Works like this:
- Build boxes with XZ's matching the aerial.Height doesn't matter - exaggerate the
height.
- Taking care not to Z-tilt the camera, shoot one height sample point of
known XZ from standing at another.
- Rotate the PhotoPoint round Y to face the target point - the target point
should appear directly above or below its position on the photo.Rotate the PhotoPoint
round X till the vertical lines on the photo match the vertical lines of the
box wall edges.
- Alter the height at the target point till it matches its Y position on
the photo.
- Having altered the ground height, ensure all photopoints are the correct
height above the ground (use a MaxScript for this!!!)
- To verify the data, repeat the process shooting the first photopoint from
the now-y-adjusted first target point.
A typical height-sample range would be between two junctions (points at
building edges or maybe pavement edges).
6-Apr-2005
4.5
Checked GetMapping for
terrain/height data.But it doesn't
look like they've got anything accurate enough to be of much use to me. And
its £50+vat!
Got the parsing system up to a
usable level.
Start-point and collision boxes
are parsing in perfectly :D
Added debug draw for the
collision system - currently just draws its collision boxes.Looks to be a frame behind somehow.
TODO: Swap vectors' Y & Z
axes when exporting using MaxScript.
Steve T was telling me about
character lighting, and the 3-point lighting they're using/trying to use on
Possession.
Showed me some really effective stuff in 'Def Jam' on PS2 which looks like
it's using ramps to force rim light effects.Might be worth bearing in mind when I come to do the zombie lighting.
Might have made a couple of
little blind changes to ZombieWorldExport.ms.
5-Apr-2005
5.75
Couldn't see a way of creating
custom modifiers (for entity classes) using MaxScript, so I'll forget that
idea for now.
Learned a big bunch of
MaxScript.The export script is
working beautifully now, writing out class properties in a nice convenient
block format.
It also writes out a class count at the beginning so that the game can
malloc stuff up.
Class properties are specified in the objects' user-defined properties.
Got the whole parsing system in
place - almost ready to roll.
Put some PD string CRC code in,
which is used by the file parsing.It's virtually the same code that B uses, which is reassuring.
Nice meaty coding session - very
pleased with how it's going.
4-Apr-2005
4.5
Learned about MaxScript; got the
script to export the level data working (custom properties and all) :)
Started setting up the
world-parsing, collision and route-finding stuff.Collision boxes are XZ rectangles from the
world file (currently a .dat).
Fairly productive evening.
1-Apr-2005
4.5
MUST keep this log checked out -
I just lost a whole entry due to sourcesafe shenanigans.
Tried out the DirectX Summer
2004 samples.
The PRT demo is eeriliy lifelike.MUST use that subsurface scattering for the zombie flesh because it
looks amazing - really fleshy.
Bought an aerial photo of
central Leamington for just under £25.Mapped it to a Xara file with a 1 metre grid and rotated it to be
grid-aligned - perfect perfect scale reference!
It covers both my walks home from work, so the area is spot-on.
Downloaded Neat Image (image
cleaning) demo - only saves jpegs.No
real use on the aerial, but should be great for texture photography later on.
Decided from looking at the
aerial that routefinding and collision want to be based on axis-aligned XZ
rectangles.
I'm wondering: should I actually
just base the town directly on the centre of Leamington??To save me umming and ahing about level
layout etc?A couple of other games
have done this and it's been really cool - going round places you've been in
real life.
Yes, fuckit, that's exactly what
I'm going to do.It should save me a
lot of time trying to decide whether stuff's realistic or not - I'll base it
all on photographs, observations and measurements.Also, it'll be a lovely souvenir of the
town I'm trying to escape from (in real life!)
Bedford Court & John Street will be the starting point for the level
creation.
Made a few decisions about
ground undulation, etc.
- I do want hills, and the
zombies will need to follow the hills.A 2D array of height, generated by MaxScript.
- I do want bowed streets, etc.
- Pavement collision:
Axis-aligned XZ rectangles should cover 90% of cases.Later, I might support non-aligned
rectangles too, because I would like non-aligned
streets if requried.Pavements will be a standard height offset
from the hills.Background zombies can
skip pavement collision (very background zombies can probably skip wall
collision too).
- Shadows:
Volumetric shadows are going to be the easiest to do.Not ideal (I'd prefer some fuzziness), but
I'm guessing they'll look fine, what with all the rain and stuff going
on.They might even be able to shade
the rain, which'll be nice.I'm open
to using old-skool blob shadows for background zombies.
Collision: Like I
said, this will all be axis-aligned XZ rectangles (might need to support
rotated ones later, see how it goes).
Need lines for fences / walls too.
First thing I'll do now is get
the wall collision boxes working in the current level, then I'll do spawn
points.After that, I'll probably
start a rough version of the proper town.
15-Jan-2005
5
Got the new anim-user system
working, yay!Checked the
MultiAnimation project as a reference for using D3DX animation
controllers.
Tested it with a horde of a thousand zombies; they initialised in a flash,
took up bugger all memory and shut-down with no problems at all!I randomised their anim speeds to make sure
they were independent.So relieved
that worked ok.
visibility functions, zombie
culling
some maths functions and stuff
Shifted cameras' projection
precalcs (projection matrix, FOV tans, etc) out into a CalculateProjection
method.Added suitable dirty flags.
Camera FOVs are now stored
halved (far more convenient).
14-Jan-2005
2
Seperated the necessary parts of
CAnimMesh out into CAnimUser (an instance of an animated mesh), and got the
code compiling again.Tomorrow I'll
approach it fresh and (fingers crossed) get it running in this new form,
which will be infinitely better.
Help seems to be working fine
again, thank goodness.Don't know what
was up with it before.
13-Jan-2005
2
Oops, got sidetracked again by
work, the band, and sheer laziness.Must resist the temptation of those comfy nights in (except
Wednesdays) and nights out.When I
need the evenings for band stuff it's fair enough, but the rest of the time I
should be following the schedule - to the letter!
TODO: consider moving Wednesdays to a night when there's music on - Monday
maybe?Could be nice.
Fixed the exit assert by
removing a RestoreDeviceObjects call from CZombieManager::Init (not sure why
it was there).
I'm now going to concentrate on
getting the required number of zombies into the game - this will show that I
know what I'm doing.Currently they
each take a fair time to initialise and take up loads more memory than they
need.
Realised that cloning meshes was a stupid idea - need more of an
Actor/ActorInstance system where multiple zombies use the same mesh data but
with different animation states.
Reconstructed the Release
configuration in order to get "debugging info" so I can break and
step through the code.Not sure what
was wrong there.Erm, maybe a rebuild-all
would have fixed it actually, oops.
18-Nov-2004
3.5
Tried to cut down on the zombie
setup time by cloning the first zombie into all the others.So complicated though - couldn't get it
working.I might have a hope in hell
if I could access MSDN and find out about the IUnknown interface.
24-Oct-2004
5
Fixed Tubby's vertex weights
(shifting the skin modifier had meant that weights got reassigned to
different verts kinda randomly).
Gave Tubby a rough zombie walk
cycle.
Added some debug and maths
functions.
Got the zombies to chase the
player.
Added placeholder headshot
detection which results in the zombie being respawned to its start pos.
Changed the storage of the
zombies to be a fixed-size array; all the mallocing gets done at the start of
the level now.
TODO: Create the first zombie
mesh then memcpy it for all the other zombies.The setup time is currently very long
setting them all up individually.
TODO: Fix exit assert.
Realised that tubby was built
facing the wrong way :(Will correct
that in Zombie mkII.
23-Oct-2004
4
Where possible, removed files
from my VSS folder that contained 'FOP3'.The work VSS database got broken recently and I was worried that I
might have caused it by cannibalising that database for Zombies.Wasn't able to remove the file called
'aaaaaaaa'.Didn't see anything too
suspicious in it though.
Learned a bit more about
skinning and animating.
Had a go at editing Tubby's
vertex weights.Got his eyes to move
with his head by linking them to his neck bone.
16-Oct-2004
8
Finished Tubby's head and
attached it to his body.He still
looks kinda short in the game, but for now I'm putting that down to his
cartoon proportions (that'll be why his eyes are below the player's eye
point).
Used a MultiRes modifier to
reduce his res while keeping crumple zones for animation.Currently about 1250 verts.
Made bones for Tubby, set-up the
IK chains and added a skin modifier.
The old
'skin-not-moving-with-bones' thing seems to make more sense now.Seems you need to have the skin modifier
selected in the modifier stack in order to see the skin deform with the
animation.
Right - to get the skinned
animation to export properly, you have to collapse all the modifiers on the
mesh, THEN add the 'skin' modifier.
And lo, the zombies did
march.I've now got my own skinned,
animating characters in the game - made by me, from scratch! :D
Now where did I put that biscuit tin?
I'm going to bail out of my
all-nighter now - I don't think I'm up to it and I should sleep ok after
tonight's results.
15-Oct-2004
3.5
Finally got SourceSafe eating
out of my hand - such a relief.The SS
project layout is neater now too - none of those stupid dummy folders.
Continued with Tubby.Had to re-do the head but this gave me a
chance to re-familiarise myself with Max (it's been a while).
I reckon once the head is
finished and joined onto the body, I'll move onto skinning & animating
him.The purpose of Tubby was just to
learn the basics of character modelling - I shouldn't actually need to finish
the tutorial (clothes, hair, etc.).
Tried creating several zombies
instead of one - worked fine!
14-Oct-2004
3.5
The reason the zombie looked
short was that the near clip plane was too far away; brought it closer now.
Added & improved debug draw
functions.
Got a nice debug game-mode which
draws the stats and handles the debug keys.Lucky me.
Fixed the mode stack issues.
Improved the fly cam.
Need to work towards animating
the placeholder zombie next.It'll
kick ass when they start wandering around.
10-Oct-2004
4.5
Got the mode stack working.Need to handle the old mode-removal problem
cases but other than that it's working nicely.
Restructured the code into game
modes, currently 'play', 'debug' and 'fly cam' - this makes things lovely
& neat.Will add the likes of
'pause' and 'front end' later.
Got the fly cam working.
9-Oct-2004
4
Decided it was time to add a fly
cam.Started adding this, but then...
Decided it would be a good idea
to set up a mode stack.For example
the fly cam will be controlled from a 'debug' mode that gets pushed on top of
the normal game mode and stops the normal game mode receiving input.The mode stack will become very useful when
I start adding front-end, pause screens, etc.
Eventually succumbed to the
temptation of the rock night at Robbin's Well - sorry!
7-Oct-2004
4.5
Been away for a long, long
time.I've now started following a
weekly schedule telling me what to work on each night (Monday & Tuesday
is overtime, Wednesday is housework, Thursday and Friday is the demo).So hopefully I shouldn't get so sidetracked
in future, but it's still going to be slow progress.
Sorted out the scale.Zombie still looks short somehow -
something to do with the projection?
Added Debug.cpp&h and some
debug draw functions.These took a
while because I had to find out about flexible vertex formats, which were
alien to me.Turns out they're very straightforward
though.
Did a bit of tidying and
re-familiarisation.
8-Jul-2004
2.5
Discovered to my dismay that my
"optimised" rotation matrix functions were actually a wee bit
slower than making each axis' matrix seperately and multiplying them!What a topsy-turvy world we live in.So I'll update matrix.cpp/h
accordingly.
Oh - and the supposedly-fast asm 'SinCos' function I got from the net
actually appears to horrendously slow - not sure what's going on there!
Figured out how to use
QueryPerformanceCounter to measure time intervals in ticks.I'm gonna be using that a whole lot.
Ressurrected some code to print
some much-needed on-screen debug text.Allowed me to check the above timer results in speed-optimised release
build.
17-Jun-2004
1.5
Continued fiddling with Tubby...
13-Jun-2004
4
Continued the Tubby
tutorial.Exported the character as
soon as he was humanoid, and loaded him into the game with no problems.
Got snagged on MS flight sim
again.Might need to un-install it or
something...
12-Jun-2004
3.5
23:16 - First "Max
moment" of the project, while working through the Tubby McChubs
character tutorial (which is going well btw).When I tried to scroll the help page Max vanished, taking with it
about an hour's worth of almost-finished hand!Must get into the habit of saving more
frequently.
8-Jun-2004
3
Got the zombie-dragon respecting
its position, y orientation and (prolly temp) scale.This is good.Man rest now.
Added a couple more steps to the
design checklist.Next one is learning
how to build and animate characters in Max - yay!
7-Jun-2004
3.5
IT FLIES!!!!!The dragon is in the game, textured, moving
and animating - WORKED FIRST TIME!!!Wicked.
Checked-in all source
immediately after getting 'SkinnedMesh' source working, at 23.09 odd.(Errors on exit though)
Got rid of the shutdown
errors.I think I'm starting to
understand the 'InvalidateDeviceObjects' / 'DeleteDeviceObjects' lark a bit
better now, so I'll fix the method names and comments soon.
6-Jun-2004
8
Tried eliminating the
frame-range overlap in the dragon's anims - didn't help.
Tried reducing the dragon to
just bones, mesh, and 'Scene_Root' dummy (ie. removing lights, points, IK
chains, etc) - this didn't help.
Giving up on the dragon for just
now, because I can't see where the problem is.I've learned a few things from it but I
think I'll learn more from building a minimal skinned test object, which I'm
about to do now.This way, I should be
able to see exactly where the problem starts (provided I can get some
animation working in the first place).
Couldn't get any animation on
the minimal test object to appear (fine in deep-exp as always).
Dragon: tried putting the bone
names *after* the keyframe blocks - no change.
Tried making sure 'scene_root'
had an identity matrix and a keyframe at the start and end of the anim - no
change.
Decided to jettison the
'MultiAnimation' code, having realised that the 'SkinnedMesh' sample loads
and animates the dragon almost perfectly!Although ultimately I'm likely to want animation blending, it's not
worth losing all my time over at this early stage - I want to start actually making this game!I'll be able to use 'MultiAnimation' as
reference later on when I add animation blending.
So I'm about to cannibalise
'SkinnedMesh' instead...
Realised that ticking the 'use
full pathname' box in the exporter eliminates the need to put the textures
next to the source (duh).
Checked-in all source before
starting the transition to 'SkinnedMesh', at about 22.44.
5-Jun-2004
4
Remembered that the textures
currently need to go next to the .exe's - dragon now textured, and ground
texture showing in release builds.
Altered the dragon's hierarchy
to get rid of the assertion failure mentioned above, but it's still not
flapping...
3-Jun-2004
1
Fiddled about trying to work out
why my dragon won't animate, but to no avail.I notice that I can't seem to change its position either.The only output I get is D3DX: Assertion failure!
(d:\builds\nt32_chk\multimedia\directx\dxg\d3dx9\anim\loadxh.cpp 1342):
D3DXFrameNumNamedMatrices(pframeRoot) == plc->cNamedMatrices
Too tired and hungover; going
home.
2-Jun-2004
2
Managed to get the dragon mesh
loading in.Currently untextured and
not animating (?), but looking healthy nonetheless - big relief.
1-Jun-2004
1
Installed Max 6.0 and - HURRAH -
managed to export working skinned animations!This is a huge relief.
31-May-2004
5
Read some more about Québecois
& French immigration.Took the
Québec immigration evaluation a few times and got different results depending
on how I interpreted the questions, so I'm not sure one way or the other
whether I'd be accepted.Much as I
hate to admit it, it's my lack of a degree that's complicating things.
France, on the other hand: very cheap to get to, no immigration hassles
whatsoever, just even more in the way of language complications.Ah well, I should be able to make this
thing work one way or another, so I'll just concentrate on making the game as
impressive as possible (and on saving money) and worry about the rest later.
Worked through some Max
tutorials - learned a bunch of stuff about Max, modelling &
animation.Made more attempts at
exporting a skinned animation - still no joy.I'm thinking I should maybe try to get hold of Max 6, since the Panda
website says something about a Max 5.1 bones bug (?)TODO: skin-up my_tut_hand.max and try to work out exactly where the problems creep in.
30-May-2004
4.5
Oops, got a bit sidetracked by
other things these past two months...
Removed ALL borrowed code,
because it was really bugging me having it there.Learned how to multiply matrices, and used
this knowledge to reconstruct MatrixRotationZY.Added a PD MatrixRotationXYZ.Implemented an easy (inefficient) version
of MatrixRotationYZX - might write it properly later.Reconstructed Vector3RotateY.
Managed to get some debug
asserty stuff working, eg. DgbBreak.
Set up a release configuration -
all seems fine, except my ground texture has disappeared (hopefully it's
nothing too sinister though).
Installed Max 5.1 service pack
1.Got some relatively promising
results with the 'soccer' demo character.I think what I should do soon is try building a placeholder character from
scratch to see exactly where it starts to go wrong.
9-Mar-2004
2.33
Improved the player run by
adding sideways lean and height-wise bob.Still rough, but it's already pretty convincing and fun, like some
kind of an urban running simulator!It's going to kick ass man!
8-Mar-2004
2.67
Installed a cracked Max 5.1 and
Character Studio 4.0 on my machine.Some of the skinned sample files load ok into Max now, but they all
still look just as screwed after export :(
Seperated the 'UtilMatrix...'
functions off into their own file, Matrix.cpp/.h; renamed them to 'Matrix...'
Added Vector3.cpp & .h for
VEC3 vector functions such as rotation.
Look-around, run and strafe are
in and working now.Tweaked the player
height and zombie height - the scale feels pretty good now; just the
pavements that are too narrow (will revise the town plan before next town mesh
revision).
Hooked up the camera's view
& projection matrices to the zombies' vertex shader, so the zombies
appear where they should in the world again.
Not a bad little evening's work
in terms of visible results - but KEEP IT UP!!
17-Feb-2004
6.8
Got the lookaround code pretty
much working - player is talking to its camera quite nicely now, but it still
needs some work.
Added some more maths
stuff."Borrowed"
a couple of matrix functions from you-know-where.I'm not happy about borrowing stuff, but if
D3D doesn't have these functions I'm damned if I'm going to go wasting a load
of time looking for them in maths books just for the sake of not
cutting-&-pasting.I mean, 3D
maths functions are common knowledge - I'm not stealing anyone's creation; I wouldn't do that.Just benefitting from their typing, that's
all ;)If I can find some
public-domain replacements, I'll use them instead.All borrowed code is marked
"_borrowed".
->Nothing borrowed from there now
I'm now using 16-byte-aligned
matrices, vectors and quaternions throughout the game (added typedefs
MATRIX44, VEC3, VEC2 and QUAT for these).On Pentium 4's, the D3DX functions are optimised for 16-byte-aligned
data.Rar!
16-Feb-2004
4.8
Got all the input setup working
very easily.The built-in controller
setup screens are working perfectly!ACE!
Started adding maths macros
& functions.
Continued working on CPlayer and
CCamera.
15-Feb-2004
7.5
Sorted out the rendering
problems I was having; got the placeholder zombie drawing properly against
the landscape.
Took me a while to work out that
the demo character ("Tiny") was about a nanometre tall - so I
altered the character vertex shader to apply a scale factor.
Stole the CInputManager class
from the 'Donuts' sample.It's a very
good example because it already uses the same controls that this will
use!
Started changing the input code to suit the game.I'm using the Direct Input FPS virtual
actions, meaning that if someone tries to play using some crazy-ass mech
controller, they shouldn't have too much hassle doing so!
Added Utils.cpp & h - for
miscellaneous handy stuff.
