Category Archives: programming

Some things I learned about OpenGL and OpenGL ES

I have done some stuff with OpenGL and OpenGL ES in the past. I generally tried to stick with fixed function pipeline; I never had good reasons to dig deeply into features provided post-GL1.4 (most of the time, GL1.1) and OpenGL ES 1.x. So the client for an MMORPG I worked on (both the 2D and 3D) did not need anything more. Nor did the simplistic unfinished spacesim. Neither did the zombie squishing game for iPhone.

For a few months now, I’ve had access to one of the “major league” game engines (which shall remain unnamed for now!), so I had to expand my knowledge a bit. I didn’t succeed as much as I should have, but here are a few interesting things I have learned.

They may be obvious, but I’ve only learned about them very recently. And it wasn’t trivial to figure them out exactly, at least based on queries I typed in Google. “What does this OpenGL feature do and how do I put it into OpenGL ES?”

Internal texture formats

We all know the incantation:

glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA, 64, 64, 0, GL_RGBA, GL_UNSIGNED_BYTE, textureData);

So I knew there was talk about the first GL_RGBA involving an “internal format”, and the second GL_RGBA and GL_UNSIGNED_BYTE specified how the textureData argument was, well, formatted. If we replaced them with GL_RGB, we’d drop the alpha from the image. But what did all that mean?

Yesterday I ran into GL_RGB8. And GL_RGBA16. And GL_RGBA4. And GL_RGB565, and more. These can be passed as the first argument, and determine how the desktop OpenGL implementation will store the texture. If you pass GL_RGB, you’re letting the implementation determine the optimal pixel component size in bits. If you get more detailed, you get more strict about the internal format you want the implementation to use for the pixels you passed.

Today I dug into a document I ran into yesterday, from 2007, which discussed that with OpenGL ES the implementation gets to decide, and tries to honor the internalFormat argument based on the format and type arguments.

GL_BGRA is an extension

On desktop, it got promoted to core OpenGL. On OpenGL ES, it’s called GL_BGRA_EXT.

Oh, and GL_HALF_FLOAT_OES does exist on OpenGL ES.

Protip: you do want to look into headers, especially the extension header (glext.h, gl2ext.h or such), whenever you can’t find if something exists and you’re sure it should.

Reverse formats

What the hell is GL_UNSIGNED_INT_8_8_8_8_REV? How does it relate to GL_UNSIGNED_BYTE and GL_UNSIGNED_INT_8_8_8_8?

This is strictly as far as I understand it! My understanding may be wrong and I would love it if someone clarified this in the comments! (For reference, the APPLE_packed_pixels extension talks about these.)

Once again, I may be wrong. Read the above links.

Alright, so let’s imagine you passed the source format of GL_BGRA in combination with the above. You’re specifying that the data you have passed has actually not been stored in the BGRA order — but in the reverse. That’s what REV means. Reverse. So you’re telling OpenGL you’re actually passing a texture stored in the ARGB order.

Notably, that doesn’t exist in OpenGL ES 2.0, so one may end up using GL_UNSIGNED_SHORT_4_4_4_4_REV. Without actually playing with that, I suspect that would mean also having to pass data with 16 bits per pixel, which kind of defeats the purpose of not having to convert ARGB textures. You could bake the data at build time, sure, but still…

Just red, just red and green

There are GL_RG* and GL_R* formats. For example, GL_RG8, GL_RG16 and GL_R8. Yip yip.

More?

Of course, I learned even more things. And I didn’t dig into everything I want to know (what’s GL_LUMINANCE and GL_LUMINANCE_ALPHA as a texture format, and why do I want to use that?).

Less than a week before I start the new job. Looking forward to spending some of the remaining time after my Wednesday trip on studying some of the interesting things such as these OpenGL features…

GSoC 2013: Final post

So, this year’s GSoC 2013 has reached the hard deadlAnd tine at 19:00 UTC (21:00 CEST).

And, I’ve managed to fix some critical bugs in the Opal-based backend of GNUstep. Huzzah!

What we now have is doublebuffering support, image support, font support, blitting from a ‘gstate’ to a ‘gstate’ (actually between their backing surfaces while taking their respective transform matrices into account). We also have an an assortment of ‘DPS’ methods.

Everything is made possible thanks to great work of Opal’s authors, especially Eric Wasylishen who helped by fixing some outstanding bugs in Opal as well as with some text rendering code in the Opal backend. He was also of extreme assistance by keeping me great company over XMPP during late hours full of frustration 🙂

Thanks also go to everyone who ever worked or contributed on GNUstep — particularly on gnustep-gui. Especially Fred Kiefer, my last year’s mentor, whose advice always proves invaluable; perhaps initially cryptic, but always invaluable.

Thanks also go to David Chisnall, my this year’s GSOC mentor who kindly welcomed everyone to Cambridge on this year’s GNUstep meeting, and tolerated my rare status reports. 🙂

Programs running under GNUstep theme under Opal backend: Ink, TextEdit, SystemPreferences.

Programs running under GNUstep theme under Opal backend: Ink, TextEdit, SystemPreferences.

Aside from the various delays caused by job interviews, summer camps and sicknesses (as well as my overall suckiness 🙂 — the most confusing thing to me was the concept of ‘gstate’. As it turns out, GNUstep loves to keep a lot of graphics state in a class derived from GSGState. It loves to be able to switch between gstates at will. It likes resetting matrix to the identity matrix whenever it wants. It likes setting custom matrices. It likes resetting the clipping path. In short — it loves to do many things that Core Graphics doesn’t want to let you do.

Which is why I got so confused when Fred Kiefer suggested that getting rid of GState may not be so easy. Core Graphics looked so logical to me and methods that gnustep-back implemented looked like they matched them so closely. Why are things in gnustep-back so complicated?

Turns out there is a good reason — switching and copying of gstates, which DPS permits and GNUstep uses in large amounts.

So in the end, I dumped the branch related to cutting out OpalGState out of my backend. Classes in the gsc/ folder implement enough logic for it to be worth keeping them as base classes for context, gstate, etc. And Eric was kind enough to implement a way to get GState out of Opal so I can save it in OpalGState class in the backend and apply to a context when appropriate. Whew.

Programs running under Gtk theme under Opal backend: Ink, TextEdit. In background: GNUstep theme SystemPreferences.

Programs running under Gtk theme under Opal backend: Ink, TextEdit. In background: GNUstep theme SystemPreferences.

Next thing that caused a lot of problems for me was doublebuffering support. First, I got stuck with incorrectly understanding the CGContextDrawImage() API. I thought it specified the source and painted at 0,0; it turned out it specified destination. And the source was specified by creating a duplicate image. (Well, not a duplicate; same image is still used, just a different cropping argument is set.)

So I used that, and failed horribly. As much as I tried to get it to work (many hours wasted shuffling the code around) — nothing. Turns out Opal, unfortunately, did NOT make use of the cropping specifier in the ‘subimage’. When that was fixed, getting doublebuffering to work was trivial.

Images were fun. During initial sprint, I got them to work — only if they were 32bit RGBA images. And only after a lot of tries. And when context’s -isDrawingOnScreen returns YES, images break. Turns out they are drawn into an offscreen context, and then blitted onscreen using -compositeGState:....

As I didn’t figure out how exactly an offscreen context is created, right now -isDrawingOnScreen returns NO. It’s a shameless lie, but until I get some time to debug what’s going wrong, it’ll stay this way. At least we get icons in System Preferences this way!

Programs running under Gtk theme under Opal backend: Ink, TextEdit, SystemPreferences.

Programs running under Gtk theme under Opal backend: Ink, TextEdit, SystemPreferences.

Final issue that I struggled for a week or two was -compositeGState:... (and its buddy -drawGState:...). Figuring out how it’s called and from where and why was messy, and the fact that it works at all by the deadline is a small miracle. How does one debug why an image is not being correctly painted into an offscreen context — for example, one for images? How does one figure out what offsets does -gui imagine I’m applying? Did I forget to apply device offset or did I forget to update CTM? Should I flip the passed .y coordinate? Is -compositeGState:... incorrectly sourcing the image from NSImage‘s backing OpalGState, or is -DPSimage:... incorrectly painting the image into that OpalGState‘s backing surface’s backing CGContext? Does the surface exist at all when -DPSimage... is painting into it? Does the surface have the backing CGContext set up? Did I correctly apply all properties to the -copyWithZone:‘d OpalGState? Maybe code is trying to call -drawGState:...? And the most important question of them all: why the heck does it not draw?!

So right now we have a mostly functional -compositeGState:... and committed. Why mostly? Well, some images don’t seem to be drawn. And as long as you scroll the NSScrollView by grabbing the scrollbar, you’ll get a correctly scrolled scrollview. (You can also try clicking on the last visible item and pressing the down arrow: it’ll work.)

Then try clicking on an arrow button of the scrollbar. Heh — turns out -gui is scrolling it in a different way. What way? I did not try figuring out yet (as I hit the deadline).

What else is verified to be missing or broken? The blinking cursor in textviews (e.g. in Ink and TextEdit). Round buttons. I have a tableview that I can’t doubleclick on without crashing the app. (Curiously, outlineview version of the same test app does not crash.) Sometimes, pen is not lifted in appropriate places — to see this, launch SystemPreferences and open the Themes panel. Popup buttons with selection that happen to have an ‘arrow’ on the right side — in SystemPreferences, open the Font panel and look on the left — will not be correctly painted when clicked on.

Aside from being generally unhappy with performance, I’m also especially unhappy with font performance; it does seem to be a sad result of doublebuffering (everything is an image, so X11 is unhappy with that).

I’m not looking forward to fixing what remains broken, but I’ll try. I’ll try. After all — we can’t have Core Animation integration without the Opal backend 🙂

All in all, this is certainly not ‘release’ quality — but it’s close to being actually usable. I’m somewhat pleased that the backend actually works, considering how many issues I had along the way, largest of which being myself and (again) not being fanatically devoted to GSOC, which is what Google intended the students to be 🙂

GSoC 2013: Text, images, delays

image by ericwa

image by ericwa; please observe ‘textedit’ info window and its preferences on the right side

Given the delay since the last update, this should be a lengthy post. Since I’m writing this at a very late hour, it won’t be.

Shortly after the last update, I added painting of images, as long as they’re 32-bit. I finally began considering the ‘gstateless’ variant of the backend, where the OpalGState class would be removed, as well as the dependence on GSContext family of classes. The idea was that all state switching could be managed by Opal. Unfortunately, Fred explained how AppKit treats GState and the idea does not map well to Core Graphics’s one-graphics-state-stack-per-context model.

Then I went to Dublin for a job interview. When I got back, I was (shall we put it mildly) highly distracted by the experience of another travel to a foreign country. After that, a two week summer camp where I almost no progress was done. A place with no good internet connectivity (except several cafe bars) would be tolerable were there a place for me to unpack my laptop as permanently as possible. Unfortunately, any such place was a computer classroom, and my accommodation was anything but appropriate for the activity. Other distractions of a summer camp also came in play, such as ‘Hey, come here, let’s work on the movie.’

Would coming back home change things? Well, after a massive headache due to low amount of sleep during the last day of camp and during the return trip, after a cold, and after a toothache, I finally got to work.

Thanks to a lot of help with Eric on the Opal side (and even a bit on the backend side, because he’s cool like that), we now have font rendering! Hooray!

Unfortunately, you may notice that the above images feature rectangles again, instead of icons. This time they’re yellow rectangles instead of red ones. Why? -compositeGState:... method is being called after -DPSimage:... was called, so we end up with a ‘not-implemented’ marker for -compositeGState:....

What’s next? Well, wrapping up as many things as possible by the end of the week, when GSOC ends.

Yep, I could and should have done far better this year. Oh well.

GSoC 2013: App content now visible!

I completely agree with you, dear reader — I don’t like it either that it took so long to get somewhere useful. 🙂

But, here’s two screenshots that are right now making me happy. (Aside from lunch, that is.)

GSoC 2013: first useful screenshot of SystemPreferences

GSoC 2013: first useful screenshot of Calculator

Red boxes are images. On the top left of SystemPreferences you can see the menu. The white blob in SystemPreferences is where WindowMaker drew its HUD to display window position while moving the window around; doublebuffering is disabled at the moment.

Calculator uses NSWindows95InterfaceStyle, so the menus are embedded inside the window. (A necessity under XQuartz.)

So, how did I get here?

Doublebuffering

I had no idea where to start from fixing the issues. So I turned to trying to write the backing image into a file. That’s easy to do with Core Graphics; but, there happened to be a small bug with Opal. Eric stepped in once again, and fixed saving to files. I’m really thankful to him for doing this so quickly.

Now that I could output files, I saw that backing images for menus were correctly drawn. What went wrong was sending things to the screen. Menus generally worked. Hooray! That was a big confirmation that something small was going wrong. Images painted on screen were incorrectly positioned, and when running under XQuartz on OS X, even sized incorrectly. (XQuartz forces a minimum window width even for menu windows.)

At some point, I figured out, what the hell – let’s check what CGContextDrawImage() specification says. Lo and behold: the argument specifies only the destination rectangle, and does not relate to the parent rectangle! The solution is to use CGImageCreateFromImageWithRect(). I ran into a small Opal bug here, which I reported to Eric; the new image still includes the original image (just like the documentation says it should), but under Opal, this means CGImageGetWidth()/Height() report original image’s values. Also, writing such a sub-image still writes out the entire image to disk.

Unfortunately, I didn’t get this to work yet. I’m still getting weird offsets, but now I’m on the right track. Currently committed version has doublebuffering disabled (which means moving anything over the window clears that part of the surface and does not cause correct redraw).

Saving and restoring graphics state

So… -DPSgsave and -DPSgrestore. Although I was aware that this is the thing to look at in case things mess up with matrices, and I spent hours and hours looking around for what went wrong (very small amount of code, but a lot of exploration), it didn’t occur to me that I, perhaps, did not implement the methods.

Eric pointed it out to me that I didn’t.

I checked, and indeed: I was going around in circles just because I did not push state to the stack where appropriate.

So I implemented these methods under names -DPSsavegstate and -DPSrestoregstate. When it didn’t help, I looked around for ideas about what else may have went wrong.

Then today, having run out of ideas, I triplechecked what I did, and the method names were incorrect. With Eric, I only discussed that state needs to be saved; I have absolutely no idea where I pulled the method names out from or why I even though they were correct.

Then, as I went to check what the method names should be, I found they were named -DPSgsave and -DPSgrestore. And they are in, for example, CairoContext — not in CairoGState. Which is perfectly reasonable: GState classes are exactly what these methods manipulate, wrappers for state of backends whose underlying libraries do not contain a state stack, or need to maintain additional state.

So having overridden these methods, called super and then passed them on to OpalGState, I finally got something sensible on screen. Hooray!

Next: Cleanup and moving DPS methods to OpalContext

Eric has suggested moving (nearly) all methods from OpalGState to OpalContext. This definitely makes sense; if there’s some state that happens not to be maintained by Opal, it probably makes sense to add the required functionality to Opal and not to the backend.

I’ll also be cleaning up the implementation before moving on to adding fonts et al. There’s leaks all over the place because I (intentionally) did not pay enough attention while researching what the problems are.

Mid-term self-evaluation for this year’s GSoC is between July 29th and August 2nd. I’ll be submitting relatively low scores for myself; I’m personally unsatisfied with the time it took me to reach this stage. I’m hoping that the worst is behind me. On the other hand, I still have to deal with fonts. sigh

GSoC 2013: Doublebuffered painting!

Thanks to a fix that Eric applied to Opal’s handling of painting CGBitmapContext->CGImage on top of another CGContext, the backend now actually paints something sensible on a doublebuffered window. Hooray!

Well, actually, it only paints something sensible on a menu window. And as soon as the backend has to paint an update, for some reason the updated stuff is painted (mostly) upside down. I’ve tested with the Gtk theme, and the in-window menus also seem to be updated upside down.

I also seem to be unable to get menu items to be wider despite various hacks when it comes to fonts. It seems that OpalFontInfo‘s -widthOfString: is not used at all, and a series of calls to -advancementForGlyph: is instead performed, with no effect on the width of the menu whatsoever.

I’ll explore further what’s wrong with both of these issues tomorrow; preparing for a job interview today will take the rest of this day.

Note: I did not commit my yesterday’s work described above, as it mostly consists of hacks that have no useful effect onscreen. Eric’s changes to Opal, combined with a change of #if 0 to #if 1 in gnustep-back’s OpalSurface‘s init method do produce some of the visible changes described above.

GSoC 2013: Drawing some actual content

Looking at the existing [...GState DPS...] methods and comparing them to the Core Graphics functions, it looked like implementing the backend will be as much of a cakewalk as I originally presumed.

Turns out it is not meant to be.

(Today’s text is a report as much as it is a way for me to contemplate next steps. It’s a noisy stream of thoughts. Sorry about that.)

All is drawn, but nothing is seen (a.k.a. doublebuffering)

The Cairo backend contains code intended to render into a backing store and then draw the content on screen when requested to do so by X. (Presumably, also to blit the content when it updates.) Theoretically, it should be trivial to add this functionality: add a new CGBitmapContext, redirect all drawing into this new context, and then blit this content on screen when requested.

When creating the surface that targets the screen, one gets a gswindow_device_t* that describes the X11 window. To tell the rest of the gnustep-back that this surface can handle its doublebuffering (although it really can’t), we set this gswindow_device_t‘s gdriver to self, and we add flags GDriverHandlesExpose and GDriverHandlesBacking to its gdriverProtocol bitmask. This is done only if the surface is not NSBackingStoreNonRetained.

(From the above description one would think by default, backing stores would be already handled elsewhere in gnustep-back. Unfortunately, although it seems like gnustep-back might be trying to handle the case where the surface does not handle X11’s Expose event nor does it handle having the backing store, it does not do so, at least not properly.)

So, I’ve tried to do it. All drawing is redirected to a CGBitmapContext, and then in -handleExpose I’m trying to blit the content on the X11-backed Opal surface. It seems that Opal currently has some issues with the following code (i.e. I get everything drawn in pale yellow):

- (void) handleExposeRect: (NSRect)rect
{
  CGRect cgRect = CGRectMake(rect.origin.x, rect.origin.y, 
                      rect.size.width, rect.size.height);

  CGImageRef backingImage = CGBitmapContextCreateImage(_backingCGContext);        
  CGContextDrawImage(_x11CGContext, cgRect, backingImage);  
  CGImageRelease(backingImage);
}

Debug output from Opal suggest it’s going through all the expected dance moves: creating that image, converting it into destination context’s colorspace, et al. I’ll poke Eric to see if he has a better idea of what’s going wrong.

An interesting behavior is that if I simply don’t redirect drawing to the bitmap context, then I get some content on screen. Despite not having an actual backing store. Maybe GDriverHandlesBacking has side effects? I’ll have to dig in more.

(Oh, and gotShmCompletion methods seemed like they might be related. No change if I implement them.)

Everything is flipped

When I first got some content on the screen, it turned out that GNUstep’s zero-zero seems to be based on top-left corner of the drawable — opposite from what Opal expects.

So as a first attempt I went in and implemented matrix functions. While things don’t get drawn in the appropriate place yet, it seems like I’m on the right track. The thing that worries me is that I’m not sure what’s the orientation of DPS matrix as opposed to the orientation of the CG matrix. Also, I’m now overriding DPS methods and not calling super, yet there’s a bunch of code in GSGState‘s implementation that makes me think this is not such a good idea. DPS functions seem to be maintaining matrix state; hopefully they don’t ever use it directly. If they don’t, then it might be safe to simply override the getter for the matrix.

To be able to reset the Opal matrix to the identity (or, near-identity, considering it’s also flipping the coordinate system), I’ve also added a method to Opal that allows this precise operation. This is an extension over what Apple provides: Apple provides absolutely no way to set the CTM (presumably, this stands for current transform matrix) directly, which makes only a little bit of sense. I don’t see a good reason this functionality is not provided to developers. (I can think of one that’s just a tiny little bit good.)

Menus have a single item

Now all I get in a menu window is a single menu item. Drawn in place of the menu window’s title.

Meaning, drawn on a completely incorrect place. I’m not really sure if this is a menu item drawn in an incorrect place, or if this is caused by the fact that -compositeGState:..., which seems to permit blitting one GState‘s backing surface’s content on top of another GState‘s content, is not implemented.

If this is so, this may prove very troublesome to implement without adding further hacks into Opal. Core Graphics seems to provide no real way to copy a portion of a context on top of another context, apart from using CGBitmapContext.

Fonts. Oooh, boy.

Cairo backend does a lot of stuff when it comes fonts. A lot, lot of stuff. I wonder why most of that is not exposed to the rest of gnustep-back, so I’ll probably be moving that out of the Cairo backend and into an additional module within gnustep-back called fonts, or something like that. Or perhaps even into the existing gsc module.

Just taking a quick look, art backend seems to be supporting just fonts with the .nfont extension (explaining Helvetica.nfont), but I may be wrong. Taking a quick look at xlib backend, and it seems to be doing something similar to the cairo backend.

It’s going to be a pain to implement fonts, especially to do it properly. I’ll have to think about how to deal with refactoring this — cairo backend is popular and accidentally breaking it won’t be looked at kindly by the community. Still, it’ll be worth it.

Unless it turns out I should have dealt with Opal. In which case Cairo backend’s implementation will stay Cairo backend’s implementation, and Opal backend will leverage Opal for its text needs.

Yes — just as the title of this section implies, I’m not sure what I’m talking about.

Other upcoming steps

Among other things, I’ll look into getting image painting to work. GNUstep does a lot of things with images, so seeing the images might help me figure out what’s going on more easily.