[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [E-devel] Putting GL-textures into Etk, Evas, bla bla... How to choke a horse with words.

Hi Rene,

The main problem we have here is that what you'd like to do is
definitely engine-specific whereas Evas tends to abstract totally the
engine layer. Anyway, with some Evas engines it should be quite easy to
use OpenGL to render 3D stuff. Here is how to do it with the
GL-X11 engine of Evas (which is obviously the most appropriate engine
to do that since it already provides the GL context).

So, with the GL-X11 engine of Evas, all you have to do basically is to
call all your OpenGL functions before glXSwapBuffers() gets called.
glXSwapBuffers() is called by the output_flush() method of the GL-X11
engine, which is called by evas_render_updates_internal(). So the
problem is that we need to be somehow notified before
evas_render_updates_internal() calls the output_flush() method, and,
for now, there is no way to do that. I have written a small patch for
Evas to add a callback called each time before output_flush() is
called. The function to set this callback is evas_flush_callback_set().
The Evas patch is joined with this email (this patch should not be
applied to the CVS repository of Evas, it's a quick'n'dirty hack!!).

Now, all you have to do is to make the calls to the OpenGL functions in
the "flush-callback". I have attached a a small code that shows how to
do it. It's totally dirty and hacky but it works quite well :) There is
just a small problem with the Zbuffer of with the backbface culling I
guess (the cube looks weird...), but it shouldn't be hard to fix.
To compile this code, you have to apply the patch to Evas, and to
compile it with:
gcc -o evas_gl_proto evas_gl_proto.c `ecore-config --cflags --libs` -lGL -lGLU -Wall

Note that the GL engine of Evas has some serious problems though. Here
is a list of the know problems written by raster some times ago:
> 1. a font with glyphs > 256x256 pixels will just break.
> 2. images that are bigger than max texture size simply won't show
> (need to build a mesh)
> 3. cards that don't support rectangular textures will end up with
> nasty scaling results when scaling "stretched" or "squashed" mostly
> in 1 dimension (mipmap issues basically when used for 2d)
> 4. scaling down even with smooth scaling is ugly (need to use
> anisotropic filtering  if possible)
> 5. doesn't support destination alpha (easy-ish to fix)
> 6. need pbuffer support implemented for the future (we need to be
> able to render to an intermediate buffer)
> 7. if the window is fullscreen glx may choose to do page flipping not
> copy from back to front buffer. in this case evas's assumption that
> the buffer was exactly as it left it when it last rendered is wrong -
> as it actually is as it was rendered 2 frames ago, not the last frame.
There is also a problem when you have several windows using the GL
engine in your program, but I think it's probably a problem with
Ecore_Evas_GL. The Glitz engine that raster seems to be writting should
fix most of these problems :)

Regards :)

On Sun, 12 Nov 2006 12:40:37 +0100,
Rene Jensen <centipede@takhis.net> wrote :

> (Sigh) After having posted on both IRC and the webforum, I get the 
> distinct feeling that this community has been worn down by daft
> requests for getting fancy window drawing using OpenGL.
> This is a serious shame because now even the smallest utterance of
> that API makes it's author look guilty. The developers forget that
> end-users asking for something XGL'ish is a far cry from application
> developers asking for their favorite API for graphics/GLGPU
> programming, which they need to make video/3D applications using
> Ewl/Etk.
> However, as an application developer, I feel that:
> 1) Etk/Edje is perfect for the job. The efficiency is the reason I am 
> choosing E if I can get away with it in the first place. I need a
> strong and themable GUI. I *LOVE* Efl's well-organized style!
> 2) The window manager and graphics lib should NOT squander my
> precious GPU cycles on needless features. If I could make a 3D in a
> console window, I'd do that.
> 3) An extra GL layer is actually another source of errors! Once I
> made a small 3D paint-program that used GL for the painting
> procedures, http://artcamilla.dk/centipede. I spend a LOT of time
> tracking down a bug in Windows where I use Nvidias desktop switcher.
> All that was drawn on my screen was a frame delayed. Turned out that
> when I switched off the desktop switcher or even turned off screen
> rotation (I have a tablet-pc), the bug dissappeared. Wasn't in my
> program after all. Incidently, the above mentioned program used my
> own OpenGL based mini-GUI. Guess why I am looking for another GUI
> with a strong support of themes ;-)
> 4) It will be hard to write any program at all in the video/3D
> industry without having support for OpenGL.
> --------------------------------
> Ok, enough ranting. This is a verbatim copy of the forum post I made
> few days ago. By the end of this mail, I have already answered some
> of it:
> --------------------------------
> I've learned that OpenGL requests falls in two camps this place:
> Those for a way to put Evas into OpenGL (for kicks), and those for a
> way to put OpenGL into Evas/Etk (for making any kind of serious
> multimedia program). This post belongs in the latter camp.
> Without in any way claiming to be a proficient E programmer, I've
> taken the liberty to scout for the possibility of drawing 3D in ETK,
> since it appears to be a non-existing feature right now. Hope I won't
> make too big a fool of myself ;-)
> Basically all that is needed to be able to draw using opengl in X is
> the creation of a GL context using GLX. The only thing required is
> that you have a handle to an X drawable (Window or Pixmap). I can see
> that in E getting this handle is less obvious, since E is unaware of
> a specific render-backend. Opposite GTK which has an X Window for
> each widget, ETK renders each element directly to a buffer, and only
> this grand superbuffer has an associated Window - if using some kind
> of X11 backend of course. If I read the code correctly, the Evas
> software-X11 backend actually renders to a pixbuf in shared-memory.
> So far so good. I traced this Drawable to a function called
> 'eng_setup', which is handed an X Drawable (amongst other things).
> Though I can't see where this Drawable is created since I don't know
> who CALLS eng_setup, I guess it would still be possible to get the
> super-drawable somehow.
> So I guess that in order to get OpenGL support in ETK, three things
> are needed:
> 1) Access the Drawable for the root Evas (if using X11/GLX) or a
> Window (if using GDI/WGL on windows)
> 2) Knowledge about where in the root-evas the widget which on which 
> OpenGL drawing should occur is placed.
> 3) Cooperation from ETK: It shouldn't paint anything to update that 
> widget and it should make it possible to emit a "paint" signal so the 
> user can emit the proper set of OpenGL commands.
> Getting 2. to work should be easy. Accessing the Drawable in 1 is not
> at all impossible either, I guess. (save the Drawable in a global var
> in some of the E libraries and get it through a function. As to 3.,
> it is more of a mystery exactly how, but perhaps somebody sees the
> light more clearly.
> Before I go on, I'd like your sentiments on this request/suggestion. 
> Presumable somebody will cry: "No way!  E has nothing to do with
> OpenGL period! This would violate the API which is totally
> architecture agnostic, you ignorant philistine", which is a perfectly
> adult way of opening this interesting discussion. Other people may
> ask why I am not simply using this ...XXX... GL-widget. I certainly
> hope that happens ;-)
> --------------------------------
> And a few more thoughts (and solutions) I have since made:
> 1) I can see a problem if Evas uses some form of buffering. Then you 
> need to pump out the data from the GL context. This may not be as
> slow as it sounds on modern architectures and if the updating of the
> cache doesn't happen every time a single bit has been changed.
> 2) A spin on the above. In case Evas buffers use that cache, it is
> going to be hard to emit some kind of "paint" signal which will be
> needed to draw opengl into the canvas. However I studied Evas' code
> and I can see that the engine-module simply get a
>     eng_image_draw (in modules/engines/gl_x11/evas_engine.c)
> which converts into a
>     evas_gl_common_image_draw (in
> modules/engines/gl_common/evas_gl_image.c)
> In the last function, the is evident that the Image being painted is 
> actually a GL texture. So a quick solution could be to use GL's
> recent Render-To-Texture mechanism (a.k.a Framebuffers) to make the
> Image being juggled with an actual 3D image rendered at users will
> (as long as (s)he remembers to tell Evas that the "image" has
> changed). Not the fastest solution, but not bad either. The
> programmer is unburdened with "paint" calls, and Evas is free to
> repaint and buffer as it sees fit.
> Responding to my own statements earlier:
> 1) We already have access to the Drawable from the program's main 
> function, which sets up the environment. I missed that because it
> seems etk_init handles it. Thus etk_init needs to open up a bit.
> 2) If you use Render-To-Texture, you don't need anything from Etk
> except access/modification possibility of it's GLuint that holds the
> texture name
> 3) - ditto -
> Caveat's:
> 1) You still need the GL CONTEXT that get's created. For some reason 
> this isn't dont in the main function, e.g.
>     bin/evas_gl_x11_main.c
> 2) You need to be able to get the GLuint of an image.
> 3) This is inherently ENGINE SPECIFIC. No sense in asking for a 
> texture-id from a cairo backend.
> --------------------------------
> And something completely different:
> I have good experience with OpenGL and a small GDI/GL + X/GL 
> cross-platform GUI behind me. I couldn't help noticing that your 
> Win32/GDI engine files are almost empty. Just ask if you need any
> tips. I might be able to help. I also have a somewhat deep
> understanding of OpenGL Framebuffer Objects and all that "weird"
> OpenGL stuff.
> Oh, and don't use the GDI way of transferring buffered graphics data
> to a window (Device Independent Bitmap). Slow like hell. Either use a
> GL context (why not?) or DirectX. I don't know anything about GDI+ or
> the new stuff from MS.
> Ahem, I should stop talking.
> Regards Centipede (Rene Jensen)
> -------------------------------------------------------------------------
> Using Tomcat but need to do more? Need to support web services,
> security? Get stuff done quickly with pre-integrated technology to
> make your job easier Download IBM WebSphere Application Server
> v.1.0.1 based on Apache Geronimo
> http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642
> _______________________________________________ enlightenment-devel
> mailing list enlightenment-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/enlightenment-devel
Index: src/lib/Evas.h
RCS file: /cvs/e/e17/libs/evas/src/lib/Evas.h,v
retrieving revision 1.92
diff -u -r1.92 Evas.h
--- src/lib/Evas.h	30 Sep 2006 10:18:32 -0000	1.92
+++ src/lib/Evas.h	12 Nov 2006 16:41:40 -0000
@@ -396,6 +396,8 @@
    EAPI Evas_List        *evas_render_updates               (Evas *e);
    EAPI void              evas_render_updates_free          (Evas_List *updates);
    EAPI void              evas_render                       (Evas *e);
+   EAPI void              evas_flush_callback_set           (Evas *e, void (*flush_cb)(Evas *e));
 /* rectangle objects */
    EAPI Evas_Object      *evas_object_rectangle_add         (Evas *e);
Index: src/lib/canvas/evas_render.c
RCS file: /cvs/e/e17/libs/evas/src/lib/canvas/evas_render.c,v
retrieving revision 1.20
diff -u -r1.20 evas_render.c
--- src/lib/canvas/evas_render.c	18 Sep 2006 09:40:28 -0000	1.20
+++ src/lib/canvas/evas_render.c	12 Nov 2006 16:41:43 -0000
@@ -68,6 +68,15 @@
+EAPI void evas_flush_callback_set(Evas *e, void (*flush_cb)(Evas *e))
+   return;
+   e->flush_cb = flush_cb;
 static void
 _evas_render_phase1_object_process(Evas *e, Evas_Object *obj, Evas_List **active_objects, Evas_List **restack_objects, int restack)
@@ -358,6 +367,8 @@
 	obscuring_objects = evas_list_free(obscuring_objects);
    /* flush redraws */
+   if (e->flush_cb)
+      e->flush_cb(e);
    /* clear redraws */
Index: src/lib/include/evas_private.h
RCS file: /cvs/e/e17/libs/evas/src/lib/include/evas_private.h,v
retrieving revision 1.76
diff -u -r1.76 evas_private.h
--- src/lib/include/evas_private.h	30 Sep 2006 10:18:33 -0000	1.76
+++ src/lib/include/evas_private.h	12 Nov 2006 16:41:55 -0000
@@ -297,6 +297,7 @@
    } output;
    int               output_validity;
+   void            (*flush_cb)(Evas *e);
    Evas_List        *damages;
    Evas_List        *obscures;
#include <stdlib.h>
#include <stdio.h>
#include <Evas.h>
#include <Ecore.h>
#include <Ecore_Evas.h>
#include <GL/gl.h>
#include <GL/glu.h>

#define WIDTH 800
#define HEIGHT 600

static int _animator_cb(void *data);
static void _flush_cb(Evas *e);

int main(int argc, char *argv[])
   Evas *e;
   Ecore_Evas *ee;
   Evas_Object *bg;
   Evas_Object *rect;
   ee = ecore_evas_gl_x11_new(NULL, 0, 0, 0, WIDTH, HEIGHT);
   ecore_evas_size_min_set(ee, WIDTH, HEIGHT);
   ecore_evas_size_max_set(ee, WIDTH, HEIGHT);
   e = ecore_evas_get(ee);
   evas_flush_callback_set(e, _flush_cb);
   /* We create a black background rect because Evas doesn't clear the color buffer */
   bg = evas_object_rectangle_add(e);
   evas_object_color_set(bg, 0, 0, 0, 255);
   evas_object_move(bg, 0, 0);
   evas_object_resize(bg, WIDTH, HEIGHT);
   /* The rectangle is here to show that Evas can render object while OpenGL is also rendering */
   rect = evas_object_rectangle_add(e);
   evas_object_color_set(rect, 0, 200, 180, 255);
   evas_object_move(rect, 10, 10);
   evas_object_resize(rect, 30, 30);
   ecore_animator_add(_animator_cb, e);
   return 0;

/* Forces Evas to redraw. By default, an animator is 30 times per second */
static int _animator_cb(void *data)
   Evas *e;
   if (!(e = data))
      return 1;

   evas_damage_rectangle_add(e, 0, 0, WIDTH, HEIGHT);
   return 1;

/* This is called before glXSwapBuffers() gets called.
 * The calls to the GL API have to be placed here */
static void _flush_cb(Evas *e)
   static float rot = 0.0;
   /* We have to disable the scissor test or sometimes it's not rendered.
    * Not sure why... */
      gluPerspective(45.0, (float)WIDTH / (float)HEIGHT, 0.1, 100.0);

      glTranslatef(0.0, 0.0, -5.0);
      glRotatef(rot, 0.3, 0.6, 0.0);
         /* Top of cube */
         glColor3f(0.0f, 1.0f, 0.0f);
         glVertex3f(1.0f, 1.0f, -1.0f);
         glVertex3f(-1.0f, 1.0f, -1.0f);
         glVertex3f(-1.0f, 1.0f, 1.0f);
         glVertex3f(1.0f, 1.0f, 1.0f);
         /* Bottom of cube */
         glColor3f(1.0f, 0.5f, 0.0f);
         glVertex3f(1.0f, -1.0f, 1.0f);
         glVertex3f(-1.0f, -1.0f, 1.0f);
         glVertex3f(-1.0f, -1.0f, -1.0f);
         glVertex3f(1.0f, -1.0f, -1.0f);
         /* Front of cube */
         glColor3f(1.0f, 0.0f, 0.0f);
         glVertex3f(1.0f, 1.0f, 1.0f);
         glVertex3f(-1.0f, 1.0f, 1.0f);
         glVertex3f(-1.0f, -1.0f, 1.0f);
         glVertex3f(1.0f, -1.0f, 1.0f);
         /* Back of cube */
         glColor3f(1.0f, 1.0f, 0.0f);
         glVertex3f(-1.0f, 1.0f, -1.0f);
         glVertex3f(1.0f, 1.0f, -1.0f);
         glVertex3f(1.0f, -1.0f, -1.0f);
         glVertex3f(-1.0f, -1.0f, -1.0f);
         /* Right side of cube */
         glColor3f(1.0f, 0.0f, 1.0f);
         glVertex3f(1.0f, 1.0f, -1.0f);
         glVertex3f(1.0f, 1.0f, 1.0f);
         glVertex3f(1.0f, -1.0f, 1.0f);
         glVertex3f(1.0f, -1.0f, -1.0f);
         /* Left side of cube */
         glColor3f(0.0f, 1.0f, 1.0f);
         glVertex3f(-1.0f, 1.0f, 1.0f);
         glVertex3f(-1.0f, 1.0f, -1.0f);
         glVertex3f(-1.0f, -1.0f, -1.0f);
         glVertex3f(-1.0f, -1.0f, 1.0f);
      glColor3f(1.0, 1.0, 0.0);
   rot += 2.0;
   /* Here, we need to restore the PROJECTION and the MODELVIEW matrices
    * previously used by Evas (I thought glPushMatrix()/glPopMatrix() should
    * have saved these matrices but it doesn't seem so. */
   glOrtho(0, WIDTH, 0, HEIGHT, -1, 1);
   glScalef(1, -1, 1);
   glTranslatef(0, -HEIGHT, 0);