This is an attempt at refactoring the filters code so I can
later implement GL support. This patch adds a few extra changes
to remove avoid calling functions of libevas from the software
engine: use the draw functions from static_libs/draw rather
than evas_common APIs.
Summary:
If the last item before ellipsis item has bigger width than its advance,
evas_common_font_query_last_up_to_pos() function can find wrong ellipsis position.
When Evas finds a position for non last item, Evas must care about additionally
available space for glyph's width of the given x position.
ex) the last item's glyph before ellipsis item has a tail to draw above the ellipsis item.
@fix
Test Plan:
Test case will added as comment.
(Becasue of font license problem.)
Reviewers: herdsman, raster, jpeg, woohyun
Subscribers: cedric, Blackmole
Differential Revision: https://phab.enlightenment.org/D4727
It seems OSMesa was recently updated to not expose symbols statically,
so dlsym() returns invariably NULL. GetProcAddress must be used. Note
though that the extension "EGL_KHR_get_all_proc_addresses" is not
present (OSMesa is OpenGL, not GLES), and there is anyway no list
of extensions in OSMesa (at the WSI level, glGetString() returns a
ton of GL extensions as expected).
My OSMesa version is 11.2.0 (mesa 17.0.1).
This fixes make check.
@fix
i found evas_common_draw_context_apply_cutouts() was procsessing 300+
cutouts and as it's O(n^2)/2 to try and merge adjacent rects for
cutouts this really performs like complete junk. we apply cutout rects
a LOT. this is not the best solution, but it's quick and much faster
than doing the clipouts which drop framerate to like 1-2fps or so in the
nasty case i say (tyls -m of photos in a dir with a 2160 high
terminal).
this figures out the target area to limit the count of rects
significantly so O(n^2) is far far better when n is now < 10 most of
the time. and for the few operations where it's a high value this now
uses qsort to speed up merges etc. etc.
@optimize
This fixes evas_object_image_save after changing the orientation
of an image in the GL engine. In SW engine the pixel data is rotated
in memory, so things worked fine from the beginning. In GL we may
have to go through loops and hoops in order to rotate and fetch the
data from the GL texture.
This should fix ce45d44.
for gl noscale buffers are texture atlases that are fbo's. the point
is never to scale or transofmr them but to render them pixel for pixel
and just store pre-rendered data where its cheaper to do this than
rebuild every time. this is the enigne infra for sw and gl with the gl
code... it SHOULD work... in theory...
preparing an object is a good idea. especially with gl. you want to do
texture uploads BEFORE using textures all in one batch. otherwise this
may mean the gl implementation has to make a copy of your data in a
tmp location then copy it in later when texture becomes "unused" as it
may be in use at the moment, or it may have to stall and wait.
i have seen somewhere around 7-10% speedups on nvidia and intel
drivers with this on given a very special test case i brewed up (1000
32x32 images where i change 1 pixel every frame). this should have
impact really when we are modifying textures a lot. this is all i've
implemented for now, but this should/would/could do much more like
re-order map, proxy renders to render FIRST in a pre-render list
instead of inline and to pre-render fbo/buffer content for complex
objects like text or textblock etc.
Some engines should using sending surface damage, until now we'd only ever
provided them with buffer damage.
The difference is that surface damage is the damage to the surface the
compositor is displaying, and the buffer damage is the damage to the
buffer the client has rendered. These are different when the client
is using multiple buffers of different ages to render into.
Anything that calls eglSwapBuffersWithDamage, wl_surface_damage() or
wl_surface_damage_buffer() should be using surface damage, and not
buffer damage.
This patch is intended to make no functional change - any flush cb that
used buffer damage before still should. Actual fixes to follow.
Apologies if I broke any engines - it's a bit of a copy and wasteland
out here.
EO is now extremely restrictive wrt. threads so that efl_data_scope_get()
can't work outside the main loop. This patch fixes the usage to create
sw buffers as shared objects (accessible from both the main loop and evas
async thread) and use plain old pointers where possible.
The buffers now have no parent because efl_add(CLASS, obj_from_mainloop)
does not work with shared objects. This is bad, as the buffers conceptually
belong to the main loop, and only need to be accessible from the draw thread
for a few calls. The main loop determines their lifecycle.
Fixes T4628
As ector objects are acessed by draw thread we need to create it as
shared object in order to access it from other thread.
Note: there is some performance lag...
Summary: make ector object as shared eo object to acess from other thread.
Reviewers: cedric, jpeg, raster
Reviewed By: jpeg, raster
Subscribers: cedric, jpeg
Differential Revision: https://phab.enlightenment.org/D4319
This brings support for the eo api for external buffers (like
the old data_set / data_get). The new API now works with slices
and planes.
The internal code still relies on the old cs.data array for
YUV color conversion. This makes the code a little bit too
complex to my taste.
Tested with expedite for RGBA and YUV 422 601 planar, both
SW and GL engines (x11).
It has been discussed on the ML (thread: "[RFC] rename efl_self") and
IRC, and has been decided we should rename it to this in order to avoid
confusion with the already established meaning of self which is very
similar to what we were using it for, but didn't have complete overlap.
Kudos to Marcel Hollerbach for initiating the discussion and
fighting for it until he convinced a significant mass. :)
This commit breaks API, and depending on compiler potentially ABI.
@feature
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.
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
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 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.
Summary:
Evas Image should be independent of render engine.
So remove native.func.data member of RGBA_Image, Evas_GL_Image struct.
And remove data argument,too.
Test Plan: Local test, Tizen3.0 mobile, Desktop englitenment
Reviewers: jpeg, spacegrapher, wonsik
Subscribers: cedric, dkdk
Differential Revision: https://phab.enlightenment.org/D3850
1. unsigned char* as a return type was not even compatible
with the default colorspace (ARGB: 32 bits).
2. Change all unsigned to int for... uh... simplicity
unsigned is more correct than int for things like width,
size or stride, but in fact having both ints (x,y) and unsigned
ints makes the code more complex.
This is a matter of personal taste.
- image_native_init
- image_native_shutdown
init() will be used to test whether the engine supports a
certain type of native image.
Note: Native image support is very much dependent on the engine,
and some stuff like opengl should work everywhere (even in sw
with osmesa) but that's not the case.
I just ran my script (email to follow) to migrate all of the EFL
automatically. This commit is *only* the automatic conversion, so it can
be easily reverted and re-run.
Summary:
Evas Text, Textblock, Textgrid keeps own language information.
This language information could be vary from the result of setlocale().
Especially, Evas Textblock supports <lang> tag. The language could be
changed in the middle of text. All of these language has to be used
for harfbuzz shaping.
@fix
Test Plan: N/A
Reviewers: herdsman, raster, woohyun, tasn
Reviewed By: tasn
Subscribers: cedric, jpeg
Differential Revision: https://phab.enlightenment.org/D3628
This implements a generic way of scaling buffers, using fake
RGBA_Image wrapping ector buffer maps. The underlying algo is
still the good old linear sw scaler.
Now the filters *should* be back to their previous level of
usability. Performance will probably be even worse than it was
before, for GL, as more glReadPixels may be involved. Optimization
now consists in actually implementing the filters with GL shaders.
This fixes crashes, adds safety, and notes a couple of things that
are not yet implemented:
- Make an Evas_GL_Image from an RGBA_Image so we can draw it on the
canvas. This means Evas.Ector.GL.RGBA_Image.Buffer
- Readable Evas_GL_Image objects with gl_read_pixels
--> Implement proper map() & unmap() for GL buffers
This is a major refactoring of the evas filters submodule.
Use Ector.Buffer and the map/unmap methods instead of directly
accessing image buffers with RGBA_Image. RGBA_Image is still
used under the hood, for two reasons:
- Required for the final output (blend onto Evas itself)
- Required for the scaling routines
FIXME:
- Breaks proxy support (ie. all kind of texturing).
- This breaks filters support for the GL engine.
Since Evas still relies entirely on Image_Entry and Evas_GL_Image,
we will need an engine-specific wrapper object creating a Buffer
around an existing cached image.
Currently only SW support is implemented. GL will be more fun to
do (with glReadPixels and whatnot).
The original plan was to have two different surfaces for GL and SW,
but this is probably not going to happen anytime soon. So, move
the implementation back to lib/ector. This avoid a file duplication.
Rename a few things:
- draw helper -> efl_draw
- Ector_Rop -> Efl.Gfx.Render_Op
- ECTOR_ bla bla -> DRAW_ bla bla (base pixel ops)
- ector_memfill -> draw_memset32 (and invert arg order to match memset)
The main rasterizer file is now draw.h in static_libs/draw
This is a non functional change, simple code refactor.
Ector Surface now inherits from Ector Buffer, and the current
two renderers (SW and Cairo SW) use Ector.Software.Buffer
implementations for pixel surfaces.
Basic pixel handling is merged and will allow easy extension
(color conversion, etc...).
Buffer classes are Mixins to be fully implemented by the final
class, such as: Ector.Software.Buffer, Ector.Software.Surface
or Ector.Cairo.Surface.
This is a large ugly commit. Sorry.
The code is quite a mess right now.
This rely on a faster code path to upload dynamic texture. Once we get support
for gbm, we should see significant performance improvement in speed, but this
first step is already a 5 times improvement (Ok, we get from really bad, to not
really useful...).
The version field was not properly set for GLES 1 and 3 (but not tested),
double free() could happen on the API structs, and empty API structs
could be returned.
Summary:
In Evas-SW-Generic/X11, native_unbind_cb is not called sometimes, although
native_bind_cb was called.
Some native surface's cases, both native_bind_cb and native_unbind_cb should
be called for mapping and unmapping, eg. with tbm_surface.
@fix
Test Plan:
Evas Native Surface with pixmap sample.
Evas Native Surface with tbm(this sample can work in Tize Device)
elementary_test
Reviewers: raster, jpeg, cedric, spacegrapher
Subscribers: JoogabYun, scholb.kim, dkdk
Differential Revision: https://phab.enlightenment.org/D3317
- image_file_colorspace_get
This will be used to determine the best loading format, especially
targeted at edje_cc / edje_decc so we can avoid ETC re-encoding.
- image_data_has
Checks whether the image is currently loaded in (CPU) memory,
and also returns the current colorspace.
so the evas thread renderer didnt START rendering until evas FINISHEd
walking all objects generating a render queue. this means all the cpu
time spend generating commands couldn't allow a parallel thread
actually go and DO the rendering.
this flushes the render thread every render command thus waking up the
render thread to work in parallel to the mainloop generating commands.
this actually means int he traces i see the render thread finished byt
he time evas_render completes thus brinign forward the frame display
by quite a bit.
thanks to evlog for pointing this out.
@fix
While OSMesa may support surfaceless contexts, we don't support
them yet in the SW engine. Instead of switching to NULL, NULL,
let's error out and do nothing instead.
Since @raster changed the behaviour of the dirty flag on images,
damages must be added to redraw the GL surface. Evas_Image checks
if it is an Evas GL surface by looking at its native surface.
But in case of SW engine, there was no native surface information
for Evas GL surfaces. Also, the OPENGL surface type was awfully
abused for OSMesa support. Luckily EVASGL surface type lets us
pass arbitrary pointers :)
This is only one step into making the software engine actually
work the same as a proper GL engine from Evas GL APIs point of view.
This is necessary for the test suite (coming next).
As spotted by @FurryMyad I inverted the logic for source_clip.
This should restore the proper behaviour while keeping my previous
fixes working. See D2940.
Double parenthesis comparison is reserved for evaluation for assignment, not just comparison.
So for example you should do ((x = 2)) and (x == 2), but not ((x == 2)) or (x = 2).
valgrind pointed this one out. we access freed memory when we dup a
context because the context CONTAINS ptrs to things like rects for
cutouts. we didnt dup these. use the proper context dup call (and
properly ref pixman color image too). this was a random bug/crash
waiting to happen and valgrind caught it. suprising it hasnt turned up
before :/
@fix
@feature
this makes the gl engine by default not do bounding box, but instead
try and smartly merge nearby update regions. this means multiple
render passes IF your drivers support buffer age, but it seems to
actually help. in my test case on nvidia drivers which support buffer
age, i saw compositor cpu overhead drop by about 30%
After clip_image_get, the old mask may be replaced by a new one,
and unref'ed, but it is later on set back as the context mask image.
Maybe it's possible that there was 0 reference and the image
got freed in between.
No idea how to test this.
@fix
Before this patch, those EGL/EvasGL functions can not work
without a current context. But EGL does not require any
current context for those to work, or at least, this should
be left to the driver to decide.
Evas GL was only able to get a pointer to the display
if a context was current.
The display pointer should be infered from Evas_GL unless
we can find a current display. EGL does not require a
context to be current in most of these function calls.
This should bring evasgl a little bit closer to EGL in terms
of behaviour (those are EGL-only extensions, btw).
Thanks @spacegrapher for the quick review
@fix
evas_gl_native_context_get is an internal function
passed around from an evas engine to evas_gl so that we can
implement evasglCreateImageForContext without exposing
any evas engine internal structure to evas_gl.
It's all a ittle bit ugly but the previous solution with
dlsym(DEFAULT) didn't work.
i was runing perf top and noticed that evas_image_load_file_data_eet(0
was being called. in fact - it was #1 on the list of functions being
called. why? it didn't make sense. i found out. just a blinking cursor
in terminology was causing the background to be unloaded and
re-loaded. the new "actually unload" changes for 1.15 made this happen
and thus we kept sucking in new data all the time even if the
scalecache already had the data - and that was the problem.
so now calcecache prepare tells you if you don't have cached data and
if you likely then have to ensure the data is loaded. this cuts down
quite a bit of work.
while i'm at it... we definitely need to clean house on the internals
of evas. a decade+ of features, mess, optimizations needs to be fixed.
i mean really house-cleaned. rewritten clenl;y re-using existing code
where appropriate.
Those two functions were doing exactly the same thing[1], which
is free an image, so this commit only attempts to simplify the code
a little bit.
[1] Actually image_map_surface_free() might even not have worked
properly with cserve2 sw (calling unload instead of close).
Evas_Image_Orient is an enum starting at 0 thus we never go below 0. Remove
unneeded checks.
modules/evas/engines/software_generic/evas_engine.c:1558:4: warning: comparison of unsigned expression >= 0 is always true [-Wtype-limits]
modules/evas/engines/software_generic/evas_engine.c:1560:8: warning: comparison of unsigned expression >= 0 is always true [-Wtype-limits]
Summary:
Used engine function for load image/data and use texture unit through
Evas_GL_Image object
Used Evas_ColorSpace format instead Evas_3D_Color/Pixel format
Added transformation matrix for adjusting texture unit coordinates in shader
Added property in Evas_3D_Texture for mark possibility get texture without atlas
(see https://phab.enlightenment.org/conpherence/54/, I suppose it will done
after this patch)
Reviewers: Hermet, cedric
Reviewed By: cedric
Subscribers: cedric
Differential Revision: https://phab.enlightenment.org/D2371
Signed-off-by: Cedric BAIL <cedric@osg.samsung.com>