To properly implement EGL_KHR_partial_update we need to know the buffer
damage before any drawing operations take place. Add a new callback to
software_generic that takes place after combining of surface damage and
swap mode when we actually have this available.
Note: This means the three copy pasta implementations of
EGL_KHR_partial_update scattered around the tree are all wrong. bummer.
If buffer age isn't present we really want to return MODE_FULL, but we were
previously returning MODE_AUTO.
On recent mali drivers this resulted in accidental partial updates leading
to incorrect rendering.
Summary:
If native surface is multiple buffer pixmap such as named pixmap, Evas should recreate eglImage everyframe.
Because DDK get the buffer ID once at eglImageCreate time.
So if internal buffer ID is changed, should recreate eglImage.
Test Plan: Wearable Tizen.
Reviewers: wonsik, raster, cedric, jpeg
Reviewed By: jpeg
Subscribers: spacegrapher, dkdk, cedric
Differential Revision: https://phab.enlightenment.org/D4211
The previous commit exposed an issue with the region test
does not take into account the scale down factor.
Not a @fix in itself, as it depends on the previous patch.
Summary:
1) BMP loader support region decoding.
@feature
2) Fix an issue what BMP loader can't decode an 16bit image with bit field
@fix
Test Plan: attached sample codes
Reviewers: cedric, jpeg, jypark
Differential Revision: https://phab.enlightenment.org/D4228
This is reverts:
e4c641ed1e
build fix
19eb7b727fbf35620a13fb65b50d3056a484360e:
glx: Fix black windows in E on nvidia
For all extension functions, we need to match with the extension itself
since GetProcAddress() can return a non-NULL value even when the function
does not exist. Drivers can do a runtime mapping depending on the
context. So, we only trust the return value of GetProcAddress() when
we know for sure that the extension exists.
Thus, if a symbol exists we will always prefer it rather than relying
on GetProcAddress().
Also, glGetString(GL_EXTENSIONS) is now deprecated so we're lucky
it still works most of the time. glGetStringi() should be used
instead. This patch changes some of the use cases, but not all.
Fixes T3030 (again)
Fixes T4288
@fix
it serems some changes broke our buffer age querying - we were using
the wrong symbol... the EXT or ARGB onenot the core one which does
work. use that first. then we get buffer age.
@fix
Not sure how this got in there, but it looks like
gbm_surface_release_buffer was being called twice on the same buffer.
This should sort out the double-free mentioned in ref T4335
Signed-off-by: Chris Michael <cp.michael@samsung.com>
Simply add support for EVASGL native surfaces.
I removed all code related to OPENGL surfaces (deprecated stuff
from old Evas GL).
Fixes T2936
Note that here are still some missing features in Evas GL
support on Wayland (at least pbuffers support is not implemented).
make coverity happy with CID 1347410 - it's technically right that the
code assumes the im ptr is valid until then as it dereferences it
until then and then stores it under pd->image but then checks for
NULL.. it would have crashed already if it were, so rmemove the
pointless check.
This fixes an issue from 19eb7b727f where the code
would no longer compile due to not finding noext_glXCreatePixmap in
the structure. This was due to the fact that this field was not
compiled into the structure if gl_gles was defined.
Signed-off-by: Chris Michael <cp.michael@samsung.com>
This fixes calls to glXCreatePixmap that would consistently
fail on nvidia >= 360.
It seems glXCreatePixmapEXT was used instead of glXCreatePixmap,
and that function always returned 0.
One could assume always using the non-EXT version of the
function should be preferred. Unfortunately, doing so for all
the other functions brings back the black windows.
I'm taking a very careful approach by doing this only for drivers
>= 360.
Fixes T3030
@fix
so our sysv shm segments were both over-permissive (nothing bad
really, just other users could read and write to/from our pixel data
destined for the screen... they could do this to x11 directly anyway
so no real issue), but be more restrictive and use 0600 as xserver
runs as root so can read/write anyway and we only want our own uid
access. but even more - fix our shm segment flushing to not keep lots
of segments floating about like a bad smell when we don't need them.
we had a cache but it wasnt flushed when it should be since async
rendering turned up. this fixes that and we're back to agressively
flushing them out when idle.
@fix
using map_bo/unmap_bo instead of gem_map_bo_gtt/gem_unmap_bo_gtt
results in a cacheable mapping and a large performance boost.
(dmabuf will still remain turned off by default for the release)
This removes the need for khronos_[u]int64_t as well as the special
typdef EvasGL[u]int64.
Hopefully this should work on all platforms (note: [u]int64_t is
used in Eina APIs, so it is already required for EFL apps).
Fixes T3200
When using a mix of vblank and pageflip events like we're stuck with now
a failed flip is going to happen sooner or later. We fix this by triple
buffering. We shouldn't turn off the custom ticker.
We need to be creating buffers with ARGB by default so that things
like Alpha/Transparency work when toggled. If we always create with
XRGB then toggling Alpha/Transparency state fails.
@fix
Signed-off-by: Chris Michael <cp.michael@samsung.com>
Summary:
Now, if image is native, do not recreate texture when image size is changed.
Recreate next native surface set time. So during the time between image_size_set and native_surface_set, native image is fragile. Fix it.
Test Plan: Local Test
Reviewers: spacegrapher, jpeg, wonsik, dkdk, raster
Reviewed By: raster
Subscribers: cedric
Differential Revision: https://phab.enlightenment.org/D4153
So ok. haha. Amusing. But this is inappropriate.
1. It's not an INTENDED effect in theme so unless the theme ius doing
it - it shouldn't happen.
2. There are places to do this and it's NOT inside an engine. There
are high level filter and other mechanisms.
3. Actually this shouldn't be even done client-side. It should be done
compositor-side when moving a window around. Knowing velocity vector
etc. is useful to a client so the protocol can stay but doing the
animation client-side is "wrong". Some windows wobble and others do
not based on toolkit? really? sure we have to live with the CSD
difference but this? The compositor can do this JUYST fine. leave it
to compositor... OR do this properly with filters client-side. e.g. a
2d displacement map with evas filters would do the job as long as it
interpolates. If you want a way of forcing ALL objects in the canvas
to redirect to an intermediate buffer then do it up there at the
canvas level. It then works in GL and software, as opposed to only GL.
Also.... this is now causing issues for users:
<memeka> hi, from
https://git.enlightenment.org/core/efl.git/commit/?id=fb76fe55a52ac212b6870f1d74470a79ea5c5246
i run EFL_WAYLAND_DISABLE_WWW=1 terminology -> but it crashes in
wayland_egl/www.c in the setup_shaders function .... HELP?
so the fun was fune until we do a release (now) and until this causes
problems for users. Back to tried and tested code.
If you want to do this... do it right at the portable layers above.
...
Revert "wayland-egl: Fix use after free"
Revert "wayland_egl: Fix redirect to texture"
Revert "evas-wayland-egl: Add www protocol handling to wayland-egl engine"
Revert "gl_common: Add API for redirecting render to texture"
This reverts commit 2760afbb0e.
This reverts commit c937248eac.
This reverts commit c67f50b40a.
This reverts commit 0f7f4b6de0.
map struct allocation was not handled right - we assumed successthen
later checked for failure with an if() after using the ptr. this
should fix CID 1353722
Fix using mkstemp directly without securely setting umask first by
making use of eina_file_mkstemp function which does set umask.
Fixes Coverity CID1357164
@fix
Signed-off-by: Chris Michael <cp.michael@samsung.com>
i've fixed almost all the eina init/shutdown pairs to do the right
thing now... except one (ecore_shutdown) with comment inline where
eo_shutdown is not called. if this is called we are in crash land.
this needs further inspection.
Coverity reports that this bpp check is actuall dead code due to the
fact that bpp can never be < 0.
Fixes Coverity CID1357144
@fix
Signed-off-by: Chris Michael <cpmichael@osg.samsung.com>
This reverts commit 3c13ef14e3.
This introduced new build breaks, especially with in-tree builds.
In particular, DIR was not defined. See T3975.
I'll make the shaders regen an opt-in rather than on by default.
So... adding the .x files to BUILT_SOURCES means they would
get deleted during clean. This was bad, but turned to be awful
when this made the check for writability fail.
See:
8a273158782b8912462e
This commit reverts:
0b69356f1b
Notes:
Please revert if this breaks the build for anyone.
I'm pretty sure the use of top_srcdir is not recommended the
way I've done it.
I hate autofoo.
My nvidia EGL driver coultn't compile some shaders because of
the error:
error C1059: non constant expression in initialization
The proposed patch at D4068 is not acceptable since it just
brings the original problem back. Instead, make sure all vec3
variables are initialized in main() or another function, not in
the global scope.
I tested all examples in evas
When an fbo pageflip finishes, we should be marking the current fbo
(one just sent) as no longer busy since it's already been pixel
blasted to the screen. Also, if we fail to flip to a given fbo during
buffer swap, then don't leave that fbo marked as busy since it is
technically not in use
Signed-off-by: Chris Michael <cpmichael@osg.samsung.com>
When kernel support for locally mmapping buffers is unavailable and
software rendering is used, dmabuf may need to fallback to shm for the
very first frame rendered.
We need to copy that frame to a shm buffer during the fallback to make
sure it's actually displayed.
Summary:
1. For converting tbm buffer from yuv to rgb, image.data should be allocated.
2. Stride should be set by info of tbm surface.
3. The number of row is height*2. but if height is odd, last row is invalid in process of _evas_video_i420.
Test Plan: Local Test, Tizen2.4 Mobile
Reviewers: spacegrapher, wonsik, jpeg
Subscribers: cedric, dkdk
Differential Revision: https://phab.enlightenment.org/D3870
Summary:
Prevents the uninitialized surface pointer to be used when
glsym_evgl_native_surface_buffer_get is NULL.
Reviewers: devilhorns
Subscribers: cedric, jpeg
Differential Revision: https://phab.enlightenment.org/D4016
@fix
Summary:
Disables GLES3 support when it cannot be determined that it is
available. Prevents to repeatly request a (missing) gles3 support and
fail.
Fixes T3795.
Reviewers: jpeg
Subscribers: cedric, jpeg
Maniphest Tasks: T3795
Differential Revision: https://phab.enlightenment.org/D4014
Summary:
In same cases(devices) can lead to mix up attributes in shaders.
@fix
Reviewers: cedric, Hermet, raster
Subscribers: jpeg
Differential Revision: https://phab.enlightenment.org/D4012
Signed-off-by: Cedric BAIL <cedric@osg.samsung.com>
Summary:
Implemented interface Efl.Gfx.Buffer functions bufer_map/unmap for Efl.Canvas3D.Scene.
Added function e3d_drawable_texture_rendered_pixels_get to module evas_gl_3d
to getting pixels from FBO. Added wrappers for functions
e3d_drawable_texture_rendered_pixels_get and e3d_drawable_texture_id_get
to have possibility call it through engine functions.
Reviewers: cedric, Hermet, raster, jpeg
Reviewed By: jpeg
Subscribers: jpeg
Differential Revision: https://phab.enlightenment.org/D3978
This reverts commit 546ff7bbba.
It seems that eo_del() is useful and removing it was creating bugs.
The issue is that the way we defined parents in eo, both the parent and
the programmer share a reference to the object. When we eo_unref() that
reference as the programmer, eo has no way to know it's this specific
reference we are freeing, and not a general one, so in some
circumstances, for example:
eo_ref(child);
eo_unref(child); // trying to delete here
eo_unref(container); // container is deleted here
eo_unref(child); // child already has 0 refs before this point.
We would have an issue with references and objects being freed too soon
and in general, issue with the references.
Having eo_del() solves that, because this one explicitly unparents if
there is a parent, meaning the reference ownership is explicitly taken
by the programmer.
eo_del() is essentially a convenience function around "check if has
parent, and if so unparent, otherwise, unref". Which should be used when
you want to delete an object although it has a parent, and is equivalent
to eo_unref() when it doesn't have one.
We need to continue to tick even if we can't find a buffer - otherwise
we'll stop updating forever.
We also need to reset buffer busy bits to 0 if we run out of buffers or
we'll never try to assign one. (The page flip handler would eventually
clear the busy bit for the current buffer, but a few lines ago we just
set that to NULL so it will never happen...)
If we don't block the current buffer from being assigned to the new
current buffer we'll end up rendering into the same buffer over and over
and never actually double buffer.
An unknown crash happened in glDrawArrays() where color_ptr is
a non-reachable pointer that looks like a valid malloc() zone.
Unreproducible issue. I can only hope this patch doesn't actually
break anything.
This cleans up rectangle and line in particular, that were not
using common code to use GLPIPES. I honestly couldn't find a
problem though.
DMABuf buffers destroy their glimage in the unbind callback, so it's
usually already gone for the free callback.
Now we test if we actually need to destroy anything.
Also, switch to GLERRV macro for error reporting - it saves us the heavy
eglGetError() unless we've built it in intentionally.
i've been gettign persistent evas test suite suite failures ffor a
while. annoying as i can't get 100% pass anymore. it's due to segv's
in evas test suite thanks to chunks of uninitialized memory in the
mesh loaders/saver code.
this fixes that
@fix
Previously events used to use class name as a prefix and ignored eo_prefix
when specified. This is no longer the case. Events follow eo_prefix by default
now. In order to get around this for classes where this is undesirable, a new
field event_prefix was added which takes priority over eo_prefix. If neither
is specified, class name is used like previously.
@feature
We used to have eo_del() as the mirrored action to eo_add(). No longer,
now you just always eo_unref() to delete an object. This change makes it
so the reference of the parent is shared with the reference the
programmer has. So eo_parent_set(obj, NULL) can free an object, and so
does eo_unref() (even if there is a parent).
This means Eo no longer complains if you have a parent during deletion.
On shutdown we need to set wl_surface to NULL to let the engine know
there's nothing to write to - however we've been setting up an entirely
new outbuf instead of just changing the old one.
The outbuf code should free the surface structure, not the dmabuf
abstraction or a use after free occurs on fallback.
Re-organize some code to make sure we don't rely on anything that may
have already been freed.
Add a wl_surface_commit() to keep the animation timer alive through
the fallback process.
This adds a separate backend to the "shm" engine that allows allocation
of buffers via libdrm that can be turned into dmabuf handles and used
with the wayland dmabuf extension.
Currently only the intel buffer manager is supported.
The benefit of dmabuf buffers is that they don't require a texture upload
like shm buffers do, and when we have plane support they can be dropped
directly into a plane without a memcpy.
Split this into two parts, one that makes the base surface, one that
calls the potential back ends.
Once the dmabuf backend is added this will allow a fallback path to
re-initialize the surface as wl_shm if dmabuf fails.
This doesn't really care what the buffers were allocated with, so
let's make this chunk of code useful to both the shm method and the
upcoming dmabuf one
Refactor evas_outbuf_setup a bit so there's only one call to
_evas_shm_surface_create.
Also makes the NULL return on unhandled rotation a lot more obvious.
Summary:
enum Evas.Canvas3D.Shade_Mode are using for choose relevant shader source code.
So renaming have a sence.
Rename evas_canvas3d_shade_mode_set/get property to evas_canvas3d_shader_mode_set/get
Rename internal fields and functions
Reviewers: cedric, Hermet, raster
Subscribers: jpeg
Differential Revision: https://phab.enlightenment.org/D3882
some font glyphs are still allocated after tyhe last gl window is
freed which means we can't make current anymore to free textures after
that. this fixes that by flushing gl texture info from the font cache
when the last gl windows are gone.
@fix
Summary:
Add post processing render function as rendering full screen quard after
rendering to texture.
Add possibility use size of current frame in shader.
Add FXAA shader source files
Reviewers: raster, cedric, Hermet
Subscribers: jpeg
Differential Revision: https://phab.enlightenment.org/D3847
As portions of this code have been derived from existing code in
Weston, we should also be including their copyright/license text to
give credit.
Fixes T3421
@fix
Signed-off-by: Chris Michael <cpmichael@osg.samsung.com>
AFILL should be used only for certain images, which are
actually native external or dynamic content images.
All normal image don't need the AFILL flag, since they should
have proper argb and alpha flag.
We're supposed to allocate a large pool at startup and use it for
resizing to save pool allocations. However, this was broken and
we ended up allocating both a large pool and a proper sized pool
every resize.
This restores correct behaviour.
eng_window_use() uses outbuf->redirect, so the clever code in
eng_outbuf_reconfigure that avoided setting it to NULL to check
if we needed to recreate the redirect was not very clever at all.
Previous redirect to texture approach broke multiple window applications
by having only a single redirect per context. This approach allows
multiple redirections.
This follows commit f10672dd74
which apparently forgot some changes in xlib and eglfs.
I hope there are no remains of the old native bind/unbind
functions. Also... I can't build everything, so I'm just hoping
this works. Note: GL and SW native images are not using the
same internal api prototypes.
Programs crashed on a segmentation fault when the last window was
closed. The eng_output_idle_flush() function was removed... but
since gl_cocoa does not properly use *_generic modules, the
output_idle_flush() function called by the render engine was
garbage (hence the segfault).
Now nothing is done... but at least we don't crash anymore.
Derek Foreman
Minkyoung Kim
Jean-Philippe Andre
jiin.moon
Stefan Schmidt
Carsten Haitzler
Tom Hacohen
Chris Michael
se.osadchy
Cedric BAIL
DaeKwang Ryu
Chris Michael
Felipe Magno de Almeida
Ji-Youn Park
Benjamin Jacobs
Oleksandr Shcherbina
Jean Guyomarc'h
Nicolas Aguirre
Daniel Kolesa
Mike Blumenkrantz
Jean Guyomarc'h