| Age | Commit message (Collapse) | Author |
|---|
|
Preparatory commit for adding an interactive scene editing server
of sorts. It'll go in a separate listing, but needs these types
as it'll operate on rkt_context_t->scenes[].
|
|
I had benchmarked this change and it showed no difference at all
on my 2c/4c i7 X230.
But having just tried it on an RPi4B where it moved the test case
from 54FPS to 60FPS, a +10% improvement, it's worth the
readability loss.
It's interesting how Intel's cleverness discourages optimizing in
ways that benefit probably *all* the competition... even when the
optimization is such a minor change in terms of effort.
|
|
Major gain comes from eliminating the cosf() from the inner loop...
There's still a bunch left on the table for moire but even just
these changes turn 19FPS into 81FPS over here for:
'--module=compose,layers=moire\\\,centers\\\=2\,moire\\\,centers\\\=2\,moire\\\,centers\\\=2\,moire\\\,centers\\\=2,texture=none' '--video=mem,size=1366x768'
|
|
This was written when module names were going to have an
@/path/to/context "handle" syntax. But instead I went the "ref"
builtin module route, with path=/path/to/context as a setting.
While it's more verbose in the settings, it "just works"
everywhere that can take a module+settings because the ref
builtin is just another module like any other.
So this TODO is referring to something that won't happen in a
"ref" builtin world.
|
|
This conditionally will end the stream on scene 99999 if
connect=off (playback mode)
When connect=on it'll just make it show an "EXIT SCENE"
diagnostic instead of the "NO SCENE" message.
Now you just stick 99999 in the rkt:scene track to end the show.
It's assumed 99999 scenes will never be needed...
|
|
This probably needs more work... it seems wrong to be bypassing
the taps altogether when dt is 0
|
|
And just maintain it as the last ticks value after rendering with
the context...
A couple modules were already doing this manually in an ad-hoc
fashion, just make it a general thing.
Updated those modules to reflect the new situation
Especially in a rkt world with modules::mixer doing fades, it
becomes common to render the same context twice in the same frame
for the blending. We need to prevent accelerated animations in
such situations. For now let's just rely on ticks in a delta-T
fashion to prevent animating the context when ticks is the same.
modules::stars in particular needs this fixed up, upcoming commit
|
|
This moves the tap updating to a function shared by rendering and
context create... so we can have a valid externally-driven tap
value before rendering a single frame if possible. (this is
important for not having spurious frames/flickers in rkt
sequences)
|
|
It's problematic getting this stuff online at render time,
because the modules end up rendering with uninitialized tap
values in that first frame.
With this change modules can get their taps on-stream at context
create, and we'll notice and do the initial rocket update just
before returning the rkt context to prime everyone on-stream.
|
|
This really needs SIMD to fly on-cpu, but this improves things
some.
Using `--module=mixer,style=fade,a_module=roto,b_module=roto\
--video=mem,size=1366x768 --defaults --go` to test:
Before FPS: 92-95 floating mostly around 94-95
After FPS: 107-111 floating mostly around 108-109
so +14.8% FPS
(2c/4t i7 X230)
|
|
It's problematic/racy to always be dereferencing the tap pointer
to access it's current value.
Running checkers:fill_module=mixer on a 2c/4t (n_cpus=4) machine
was quite effective at crashing in mixer::render_fragment()
due to dereferencing a NULL snapshot fragment.
The "tween" T value is being used to indicate when interpolation
of the snapshots is necessary. But repeatedly re-reading the T
value via the tap pointer would race with the driving tap. In
the case of checkers, which is a threaded module, the mixer
contexts are all rendering in parallel at the same path so
they're sharing a single pipe for their T taps.
This situation of sets of contexts sharing a single path,
resulting in their taps all landing on a single pipe, is still
largely up in the air and might be actively prevented in the
future. But until then, it can be made far less crashy and
problematic by just being more careful about dereferencing the
tap to access its latest value just once at the prepare frame
stage, storing it in the local T variable in the context. Then
all the render_fragment() accesses can at least find a stable
value in the context from prepare-to-render, so at least the
snapshots are there when they should be according to the T value
etc.
Something like this will probably need to be done regardless of
what happens with the context sets sharing the same path. Since
when a given tap isn't the driver, it still has to take care to
just grab the updated value once for the frame... The tap API
can't really automagically do that single update of the local
variable when passenger though, since it's been deliberately kept
devoid of all the type-specific accessor junk (ptr/elems etc are
all void **/void* in the tap api side). Hence why this commit is
just ad-hoc updating the local variable in the else branch; it's
best positioned to do so directly. But there still needs to be
more consideration for thread-safety.
|
|
Maybe earlier versions used the absolute coordinates in the
frame, but the current code doesn't make use of this and simply
needs to confine itself into the WxH of the fragment.
|
|
This is not optimized at all and tends to hurt the FPS
significantly. This is one of those things that would hugely
benefit from SIMD, but even without SIMD it could be done better.
I just slapped together something obvious for now, as I'd like to
focus more on the rkt side but need a better fader for scene
transitions than style=flicker.
Also changed {a,b}_module= preferred values to blank,compose
so you see something happen if you just run --defaults.
Otherwise, the compose,compose would just fade between two
identical compositions invisibly.
|
|
Originally I was thinking a variadic inputs= like compose::layers
would be desirable, with the T value's integer serving as an
index into the inputs, and the fraction between the integers as
the mixing T.
But I changed my mind and am instead constraining mixer
explicitly to two distinctly named modules; a_module= and
b_module= with the T value 0-1 mapping to a-b.
|
|
Not sure why this was -ENOMEM, module lookups don't allocate
anything...
|
|
I thought the build was already using -Wall but that seems to not
be the case, maybe got lost somewhere along the line or messed up
in configure.ac
After forcing a build with -Wall -Werror, these showed up.
Fixed up in the obvious way, nothing too scary.
|
|
Similar to fill= and fill_color=, these influence how to treat
the "clear" cells.
Until now "clear" cells would unconditionally just be cleared via
til_fb_fragment_clear(). Now they can also be filled
w/color,sampled,textured, and maybe in the future a clear_module=
will also be introduced.
Note that for now I've left the clear_color randomizer disabled,
see code comments.
Having this setting also makes me wonder if the "filled" cells
should be optionally cleared using the "clear" setting first.
Imagine a scenario where you have fill_module=shapes, and you
want the color around the shape to match the clear color, for
instance... Especially once you can pass a color down to the
fill_module, controlling these things becomes more critical.
There's definitely more work to do here.
|
|
Preparatory commit for introducing equivalents for the "clear"
cells.
This renamed the color= setting to fill_color=, in addition to
the internal naming.
|
|
This probably needs more work, but this at a minimum should
prevent us from leaking contexts in the stream at the myriad
paths we construct them at.
Context registration replaces what's at the existing path, but
rtv produces all sorts of setup paths, and I haven't added any
explicit unregistration of contexts at this time. That might
change, but for now let's just use this gc mechanism even if it's
a temporary hack.
|
|
this is very early/unfinished, hence experimental flag
|
|
The modules don't have to defend against this, vestigial from
simpler times
|
|
compose must have been derived from rtv originally, which uses
txt.h.
|
|
a390e82 stopped using this, but didn't remove it.
As it was initialized to NULL, it was deffectively all a NOOP.
|
|
The return value was just being ignored previously, and that
really starts mattering in a world with contexts finding others
by user-supplied paths making such failures far more likely.
|
|
These are already reality as of late
|
|
Just a boring replacement of the ad-hoc n_cpus context creates
for the fill_module contexts with the newly added libtil
equivalent function.
Future commits will expand the libtil side to get module contexts
registered for discovery purposes on-stream. This change moves a
bit closer towards that goal...
|
|
Somewhere along the line this leak was created, there's been a
lot of activity surrounding this stuff so it's unsurprising.
A little janky surrounding the conditional on snow module, but
that's just how snow is handled today - it doesn't get randomized
like the other channels do.
|
|
Rather than creating an orphaned settings instance private to
til_module_setup_randomize(), the function now requires the
settings instance be provided.
The one remaining caller of this function is modules::rtv. Now
that rtv is responsible for creating the settings beforehand, and
the settings may be created with a path prefix, rtv gets its
til_module_context_t->setup.path prefixed for all the channel
settings.
Another improvement is now the channel settings instance gets
created from the module name as the settings string. So while
it's not yet possible to sparsely specify settings with others
being randomized, at least now when log_channels=on is in effect,
the printed args include the top-level channel module.
Having proper complete paths for the rtv channel modules is
especially visible in --print-paths output FYI.
An interesting test for exercising all this is:
```
$ src/rototiller --module=rtv,duration=0,context_duration=0,snow_module=none,channels=all,log_channels=on --print-pipes --defaults --go 2>/tmp/channels
in another terminal:
$ tail -F /tmp/channels
```
watch the chaos unfold
|
|
Montage would randomize orphaned setting instances for the
participating modules @ context create time.
This not only produced montage tiles one couldn't configure via
settings even if they wanted to, but it also produced partial
paths due to the orphaned settings instances.
With this commit montage tiles are configurable in the same way
compose::layers are; a comma-separated list of modules with
settings accompanying them.
Randomizing is no longer performed, but if seen via something
like rtv, that randomizer will operate on the regular setup
machinery to produce randomized montages.
One new ability delivered with tiles= is you can specify the same
module repeatedly to produce a tiled display of the same thing.
Those instances may have the same or different settings, it's
totally controllable.
This also opens up the future for more interesting things like
shiftng ticks in the montage tiles... imagine showing the same
module a few times in each row, but offsetting ticks into the
future/past in the columns. For ticks-driven modules, you'd see
the future/past frames side by side, like a flipbook effect.
This leaves rtv as the only til_module_setup_randomize() caller
remaining...
|
|
Once til_module_context_t.module was introduced, this vestigial
module member @ compose_layer_t.module became redundant.
So here it's dropped in the obvious manner, but the
compose_layer_t struct is retained despite only having
til_module_context_t* now. This is in anticipation of future
additions where compose settings may set per-layer/texture
rendering behaviors (think alpha, colors, texturing toggles, etc)
|
|
This drops the seq_module= setting in favor of a scenes= setting
in the same style of compose::layers; a nested settings instance
composed of more nested settings instances, one per scene.
A nice side-effect of this change is it no longer uses
til_module_setup_randomize() at all, which was being used to mix
up the seq_module's settings in a pre-nested-settings world.
A new Rocket sync track is introduced named "$context_path:scene"
for selecting which scene to render.
For now all scenes get created @ context create time, and persist
for the entire rkt context lifetime. In the future the context
lifetimes might become explicitly controllable with separate
Rocket tracks used as booleans. This becomes relevant once
modules can make use of existing contexts located within the
stream at their respective context paths. Something necessary
for integrated transitions between scenes using stuff like
fade modules which haven't been added yet.
With this change you can already enumerate a set of scenes in the
rkt settings string, each 100% explicitly configured, and have
Rocket track data select which scene to render on the timeline,
and manipulate the taps at their scene-specific
context-path-derived track names.
In addition to the need for modules picking up existing contexts
on the stream, rkt probably needs a way to interactively
add/remove/modify scenes then spit out the serialized settings
string for the current state of the world.
As these aren't functionalities provided by GNU Rocket, and it's
unclear how receptive upstream GNU Rocket/glrocket maintainers
would be to such additions, rkt will likely first add another
listener for a strictly scenes-editing client to connect
alongside the GNU Rocket stuff. Just something that shows the
current scenes table, and provides a way to edit/add/remove rows
there, with the changes realized in rkt real-time. Then the
Rocket Editor will just continue using the rkt:scene track to
numerically index into this scenes table, without the Rocket
Editor having any visibility or awareness of what's going on in
that table. Probably ok as an initial stab at making demos with
this stack.
|
|
Mechanical change removing some rkt_setup_t* casting verbosity in
rkt_create_context()
|
|
This eliminates the ad-hoc track_name[] allocation and
construction, since the track_name wasn't being used after
getting the track anyways. No point wasting the memory on it,
and the little helper constructing the name on-stack exists now
for another future use @ rkt_create_context().
|
|
Preparatory commit for adding scenes and a $rkt_setup_path:scene
track for selecting them. This will also likely replace the
whole track_name allocation/construction in rkt_pipe_t since we
don't actually make use of that name after getting the track
(except maybe for debugging purposes)
|
|
Mechanical rename just to make this consistent with
til_module_setup()/til_module_setup_finalize()
I should probably do a cleanup pass throughout the til APIs to
standardize on a subject-verb-object or subject-object-verb
order... Things have become a little inconsistent organically
over time
|
|
This changes til_setup_t* from optional to required for
til_module_context_t creation, while dropping the separate path
parameter construction and passing throughout.
|
|
This replaces the few ad-hoc til_module_t.setup() setup-baking
callers with the new til_module_setup_finalize() which always
produces a til_setup_t having an appropriate path, even when
there is no til_module_t.setup() method.
|
|
This commit adds passing the settings instance to til_setup_new()
which is used for deriving a path for the setup via
til_settings_print_path() on the supplied settings.
That path gets an allocated copy left in the returned
til_setup_t at til_setup_t.path
This path will exist for the lifetime of the til_setup_t, to be
freed along with the rest of the baked setup instance when the
refcount reaches 0.
The incoming til_settings_t is only read @ til_setup_new() in
constructing the path, no reference is kept. Basically the
til_settings_t* is just passed in for convenience reasons, since
constructing the path needs memory and may fail, this approach
lets the existing til_setup_new() call error handling also
capture the path allocation failures as-is turning
til_setup_new() into a bit more of a convenience helper.
Note that now all code may assume a til_setup_t has a set and
valid til_setup_t.path, which should be useful for context
creates when a setup is available.
|
|
Commit 7c8086020 switched the til_setup_new() api to support NULL
free_func for free().
This mechanical change pivots to that instead of the awkwardly
cast free() parameters.
|
|
When voronoi is overlayed the colors get sampled, so randomizing
them is pointless work every frame.
|
|
This seems to work on my 2c/4t laptop and is certainly faster.
But I'm not being careful about using atomics loading/storing the
d->cell pointer, which seems problematic. Surprisingly things
aren't crashing here despite that, maybe on a non-x86 smp box
it'd be a different story.
|
|
Preparatory commit for doing the voronoi distance calculations in
parallel when possible, as part of render_fragment() instead of
all in prepare_frame().
Not all of the distance calculation work can easily be threaded,
but it should be possible to compute the post-seed distances
concurrently within the spatial bounds of the tiled fragment.
This commit doesn't actually change anything functionally, as
it's just splitting the old voronoi_calculate_distances() into
two and calling them both in succession still from
voronoi_prepare_frame().
Subsequent commits will work towards making the render_pass()
fragment-aware then ultimately moved to voronoi_render_fragment()
|
|
This variant is kind of a broken hack, and its brokenness becomes
more apparent in a threaded voronoi world.
So just drop it for now.
I am interested in more voronoi variants, but they can't
compromise correctness/introduce instabilities or significantly
interfere with performance improvements like threaded rendering.
The dithered-ish look of dirty=on was an interesting variant
though... bummer.
|
|
trivial change; voronoi_sample_colors() only reads from the fragment
|
|
Like modules/checkers required for fill_module, we need to do the
same for for rtv::snow_module.
There's more work to do on rtv::channels, but that's still
unsettled stuff in terms of settings syntax since rtv randomizes
settings. It's desirable to have the rtv settings able to
specify which settings to hold constant at a specific value
per-channel, leaving everything else for randomizing on channel
switch.
But there's no syntax for that kind of stuff currently, and it
seems like there's a need to communicate during the setup_func
dance when we're in a "settings optional because we'll fill them
in automatically at time of use later" to the front-end.
It's not strictly a front-end issue though - because the back-end
setup_func actually controls the forward progress. From the
current setup_func's perspective, everything's important to it
and must be fulfilled. And we certainly want the setup_func to
continue informing the setup process.
So it's more like the channel settings being populated via rtv
still need to all get populated, rtv just needs a way to add an
attribute to mark which settings are static vs. which should get
randomized on every use.
Perhaps there should just be a special value syntax reserved for
saying "random value" and the front-end can apply that, but then
a til_module_randomize_setup() could detect that too in a
per-setting flag the front-end set. That way the value gets
re-randomized, while the ones without that value set get left
alone.
Yes, I know this isn't the appropriate place for such commentary.
But nobody is reading these things anyways on my toy side project.
|
|
|
|
This is harmless as long as rtv stays hermetic.
But if rtv gets used in a composite scenario in the future by
either removing the hermetic flag, or allowing forced overrides,
n_cpus=0 will cause the nested module contexts to become threaded
on SMP machines.
That's problematic if the outer module context is already a
threaded render. What's appropriate here is to just propagate
the n_cpus down so if an upper layer has already gone threaded,
it will be sending down n_cpus=1 to serialize the nested
instances.
In practice, as-is, this change basically changes nothing, but
prepares for a potential future where rtv participates in
threaded compositions.
Through a lens of "rtv just rejiggers scenes and there settings
on a timer from a settings-specified subset of modules and
settings" it's arguably useful as just another module. Sometimes
you want something to change itself up periodically in say a
compose layer.
So preparing for this possibility isn't really all that
far-fetched/hypothetical.
|
|
Currently settings instances get labels from three sources:
1. explicitly labeled by a root-level til_settings_new() call,
like main.c::til_settings_new(NULL, "video", args->video);
2. implicitly labeled in a spec.as_nested_settings w/spec.key
3. positionally labeled in a spec.as_nested_settings w/o spec.key
But when constructing setting/desc paths, using strictly these
settings instance labels as the "directory name path component"
equivalent, leaves something to be desired.
Take this hypothetical module setting path for example:
/module/layers/[0]/viscosity
Strictly using settings instance labels as-is, the above is what
you'd get for the drizzle::viscosity setting in something like:
--module=compose,layers=drizzle
Which is really awkward. What's really desired is more like:
/module/compose/layers/[0]/drizzle/viscosity
Now one way to achieve that is to just create more settings
instances to hold these module names as labels and things would
Just Work more or less.
But that would be rather annoying and heavyweight, when what's
_really_ wanted is a way to turn the first entry's value of a
given setting instance into a sort of synthetic directory
component in the path.
So that's what this commit does. When a spec has .as_label
specified, it's saying that path construction should treat this
setting's value as if it were a label on a settings instance.
But it's special cased to only apply to descs hanging off the
first entry of a settings instance, as that's the only scenario
we're making use of, and it avoids having to do crazy things like
search all the entries for specs w/.as_label set.
It feels a bit janky but it does achieve what's needed with
little pain/churn.
|
|
Like modules/checkers required for fill_module, we need to do the
same for for compose.
It's a little more weird in compose since compose::layers is a
nested settings full of unnamed nested settings.
But compose::texture is analogous to checkers::fill_module.
|
|
The bare-value value_as_nested_settings.entries[0] setting which
serves as the name for module lookup is in a sort of no-mans land
between checkers and the underlying fill_module's setup.
So we have to do this little bit of rigamarole in checkers, being
the entity wiring up the nested module. The fill_module's
setup_func won't be doing anything to describe the name's
setting as it's only interested in its own settings.
There will likely be some helpers made later to streamline this
process of composing module/settings hierarchies.
|
