Instead of iterating through all the bits, use `__builtin_popcountl()`, provided
by gcc, which should be considerably more efficient.
Fixes#27.
Signed-off-by: Gergely Nagy <algernon@keyboard.io>
Instead of `getKeyswitchStateAtPosition`, which is long, unintuitive and feels
wrong too, introduce `isKeyswitchPressed`, shorter, better, more
reasonable (because it returns a bool - we support only two states anyway!).
Additionally, add `pressedKeyswitchCount()`, which returns the number of key
switches pressed.
Signed-off-by: Gergely Nagy <algernon@keyboard.io>
Move `KEY_INDEX` here, and turn it into a `constexpr` function, `keyIndex`, and
convert the `RxCy` macros to constexpr values instead.
Signed-off-by: Gergely Nagy <algernon@keyboard.io>
Implement `getKeyswitchStateAtPosition`, a hardware-agnostic way to peek into
the keyswitch state. Also transition the `RxCy` macros to `KEY_INDEX`, to make
it easier for hardware with more keys than 64 to implement them, and to make
their values unique across a keyboard, not just across a keyboard half.
Signed-off-by: Gergely Nagy <algernon@keyboard.io>
These functions can be used to detach from the host, then re-attach, possibly
with different properties, without having to reboot the device.
Signed-off-by: Gergely Nagy <algernon@keyboard.io>
When a keyswitch has been off in the previous cycle and is still off now, do not
call `handleKeyswitchEvent` on it. As an extension, when the whole column is
idle, skip the whole thing.
In practice, handling fully idle keys is not useful. There are many plugins
which explicitly look for this case and return early, because it isn't an
interesting event. As such, not calling the event handler in this case makes
sense, as we save not only a few needless checks in plugins, but our performance
improves greatly too.
Signed-off-by: Gergely Nagy <algernon@keyboard.io>
In `actOnMatrixScan`, there is no need to use temporary variables, we can just
pass the data directly to `actOnHalfRow`, and doing so makes the code easier to
follow.
In `actOnHalfRow`, we can further optimize things if instead of reading the Nth
bit, we always read the first, and shift the byte at the end.
All of these optimizations were done by @obra, I just wrote the commit message.
Signed-off-by: Gergely Nagy <algernon@keyboard.io>
When there are no state changes, and no keys pressed on a row, instead of
iterating through a byte bit-by-bit, just fire idle events without checking the
bits. In all other cases, do the bit-walking like we did before.
The reason this is useful is because bit-walking is costly, and slow. If we can
avoid that, we win quite a lot of performance. Since rows being idle is the most
common case on a keyboard, this is a huge net win. Even in the worst case, where
no rows are idle, this is just one byte comparison and a branch slower than our
previous implementation.
As part of this optimization, `actOnHalfRow` was lifted out into its own
function, to reduce code duplication.
Many thanks to @gedankenexperimenter for the original idea!
Signed-off-by: Gergely Nagy <algernon@keyboard.io>
Instead of pulling it in from the user sketch, do so from the hardware plugin.
The hardware and the adaptor are in close relationship anyway, and with
tweakable knobs, we do not need to use a different adaptor library in advanced
cases, either.
Signed-off-by: Gergely Nagy <algernon@keyboard.io>
This doesn't change behaviour at all; it's just a different way to do the computation,
which I think is much clearer. I also added an explanatory comment.
* It's now all bitwise operations, without arithemetic thrown in.
* It uses the same exact formula for finding bits on both sides of the keyboard.
* It saves 14 bytes in program memory.
Since we have a nice keydata_t union, use that, to convey it even more clearly
that we are mimicking the hand states in a way.
Signed-off-by: Gergely Nagy <algernon@madhouse-project.org>
By not using 32-bit ints, we already saved a noticeable amount of space. If we
follow the `*HandState` bit layout more closely, we can shave off some more
bytes too.
Signed-off-by: Gergely Nagy <algernon@madhouse-project.org>
Dealing with 32-bit numbers on Atmega32u4 is very costy, so lets try to avoid
that, and use four 8-bit uints instead. This save us about a hundred bytes of
progmem.
Signed-off-by: Gergely Nagy <algernon@madhouse-project.org>
The `SCANBIT` macro was not using `row` and `col` properly: if either was
anything else than a number (such as `col - 8`), the macro did not expand them
correctly, preserving operator precedence. As such, the right-hand side SCANBITs
were broken when used with masking, because the masking code uses `SCANBIT(row,
col - 8)`, and the `(7 - col)` part would expand to `(7 - col - 8)` which is
very different than `(7 - (col - 8))`.
This patch addresses the issue.
Signed-off-by: Gergely Nagy <algernon@madhouse-project.org>
There are situations when one wants to ignore key events for a while, and mask
them out. These newly introduced functions help do that.
They are in the Hardware plugin, because this is where it is most efficient to
implement the masks: the hardware library knows how many bits it needs, and how
best to represent the masks. We use a 32-bit bitmap here, other keyboards may
use a different size, or an entirely different approach too.
This is one part of the fix to address keyboardio/Kaleidoscope#150.
Signed-off-by: Gergely Nagy <algernon@madhouse-project.org>
Given a `row`/`col` coordinate, this new function returns the index of the LED
under that key. To be used with the `led_set_crgb_at(i, crgb)` function.
Signed-off-by: Gergely Nagy <algernon@madhouse-project.org>
Instead of doing gamma correction when setting a LED color, do it elsewhere,
when talking to the LED hardware. This way, both the programmer, and the
end-user will deal with the unaltered, raw RGB values, and neither has to care
about when and how gamma correction is applied.
The counterpart of this will be in KeyboardioScanner, that re-introduces gamma
correction on that end.
Signed-off-by: Gergely Nagy <algernon@madhouse-project.org>
If we start with isLEDChanged unset, then after flashing, the `R0C0` key will
remain lit, because we think it is off, while it is not. Setting this flag
results in the first update cycle updating everything, no matter what, which is
the safest thing we can do.
Signed-off-by: Gergely Nagy <algernon@madhouse-project.org>
Instead of using the supplied linear color component values as-is,
convert them to gamma-corrected, non-linear values first. This way, we
can use numbers like 127 to mean half brightness, and have it
automatically translated to the correct, gamma-corrected value.
Table copied from:
https://learn.adafruit.com/led-tricks-gamma-correction/the-quick-fix
This adds a bit over 256 bytes of code, but makes working with colors a
lot easier.
Signed-off-by: Gergely Nagy <algernon@madhouse-project.org>
Drop the `load_primary_layer` and `save_primary_layer` methods, because
`save_primary_layer` is not used anywhere, and as such, the whole thing is
pointless at this time.
Furthermore, if we want to allow plugins to implement EEPROM storage, then its
best if we leave the default layer save/load to the plugins too.
Signed-off-by: Gergely Nagy <algernon@madhouse-project.org>
To further improve the LED performance, sync only when there is a change. We do
this by tracking when change happens, assuming everyone uses the provided
accessors.
While we do a bit of extra work each cycle to do the tracking, that pales in
comparison to what we gain by not having to transfer data needlessly.
Signed-off-by: Gergely Nagy <algernon@madhouse-project.org>
Gergely Nagy
Jesse Vincent
Gergely Nagy
Florian Fleissner
Michael Richters
Jesse Vincent
Javaru
Gergely Nagy