2015-10-27 02:32:37 +06:00
|
|
|
/*
|
2015-10-27 04:54:46 +06:00
|
|
|
|
|
|
|
Note for ErgoDox EZ customizers: Here be dragons!
|
|
|
|
This is not a file you want to be messing with.
|
|
|
|
All of the interesting stuff for you is under keymaps/ :)
|
|
|
|
Love, Erez
|
|
|
|
|
2015-10-27 02:32:37 +06:00
|
|
|
Copyright 2013 Oleg Kostyuk <cub.uanic@gmail.com>
|
|
|
|
|
|
|
|
This program is free software: you can redistribute it and/or modify
|
|
|
|
it under the terms of the GNU General Public License as published by
|
|
|
|
the Free Software Foundation, either version 2 of the License, or
|
|
|
|
(at your option) any later version.
|
|
|
|
|
|
|
|
This program is distributed in the hope that it will be useful,
|
|
|
|
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
|
|
GNU General Public License for more details.
|
|
|
|
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
|
|
along with this program. If not, see <http://www.gnu.org/licenses/>.
|
|
|
|
*/
|
|
|
|
|
|
|
|
/*
|
|
|
|
* scan matrix
|
|
|
|
*/
|
|
|
|
#include <stdint.h>
|
|
|
|
#include <stdbool.h>
|
|
|
|
#include <avr/io.h>
|
2016-07-04 22:13:41 +06:00
|
|
|
#include "wait.h"
|
2015-10-27 02:32:37 +06:00
|
|
|
#include "action_layer.h"
|
|
|
|
#include "print.h"
|
|
|
|
#include "debug.h"
|
|
|
|
#include "util.h"
|
|
|
|
#include "matrix.h"
|
2017-08-24 08:29:07 +06:00
|
|
|
#include QMK_KEYBOARD_H
|
2015-10-27 02:32:37 +06:00
|
|
|
#ifdef DEBUG_MATRIX_SCAN_RATE
|
|
|
|
#include "timer.h"
|
|
|
|
#endif
|
|
|
|
|
2016-07-06 10:24:31 +06:00
|
|
|
/*
|
|
|
|
* This constant define not debouncing time in msecs, but amount of matrix
|
|
|
|
* scan loops which should be made to get stable debounced results.
|
|
|
|
*
|
|
|
|
* On Ergodox matrix scan rate is relatively low, because of slow I2C.
|
|
|
|
* Now it's only 317 scans/second, or about 3.15 msec/scan.
|
|
|
|
* According to Cherry specs, debouncing time is 5 msec.
|
|
|
|
*
|
improve ergodox ez performance
With these changes, the ergodox ez goes from 315 scans per second
when no keys are pressed (~3.17ms/scan) to 447 (~2.24ms/scan).
The changes to the pin read are just condensing the logic, and
replacing a lot of conditional operations with a single bitwise
inversion.
The change to row scanning is more significant, and merits
explanation. In general, you can only scan one row of a keyboard
at a time, because if you scan two rows, you no longer know
which row is pulling a given column down. But in the Ergodox
design, this isn't the case; the left hand is controlled by an
I2C-based GPIO expander, and the columns and rows are *completely
separate* electrically from the columns and rows on the right-hand
side.
So simply reading rows in parallel offers two significant
improvements. One is that we no longer need the 30us delay after
each right-hand row, because we're spending more than 30us
communicating with the left hand over i2c. Another is that we're
no longer wastefully sending i2c messages to the left hand
to unselect rows when no rows had actually been selected in the
first place. These delays were, between them, coming out to
nearly 30% of the time spent in each scan.
Signed-off-by: seebs <seebs@seebs.net>
2017-11-19 04:11:26 +06:00
|
|
|
* However, some switches seem to have higher debouncing requirements, or
|
|
|
|
* something else might be wrong. (Also, the scan speed has improved since
|
|
|
|
* that comment was written.)
|
2016-07-06 10:24:31 +06:00
|
|
|
*/
|
|
|
|
|
2015-10-27 02:32:37 +06:00
|
|
|
#ifndef DEBOUNCE
|
|
|
|
# define DEBOUNCE 5
|
|
|
|
#endif
|
|
|
|
|
|
|
|
/* matrix state(1:on, 0:off) */
|
|
|
|
static matrix_row_t matrix[MATRIX_ROWS];
|
2018-11-20 20:55:35 +06:00
|
|
|
/*
|
|
|
|
* matrix state(1:on, 0:off)
|
|
|
|
* contains the raw values without debounce filtering of the last read cycle.
|
|
|
|
*/
|
|
|
|
static matrix_row_t raw_matrix[MATRIX_ROWS];
|
Change to per-key eager debouncing for ErgoDox EZ.
Empirically, waiting for N consecutive identical scans as a debouncing
strategy doesn't work very well for the ErgoDox EZ where scans are very
slow compared to most keyboards. Instead, debounce the signals by
eagerly reporting a change as soon as one scan observes it, but then
ignoring further changes from that key for the next N scans.
This is implemented by keeping an extra matrix of uint8 countdowns, such
that only keys whose countdown is currently zero are eligible to change.
When we do observe a change, we bump that key's countdown to DEBOUNCE.
During each scan, every nonzero countdown is decremented.
With this approach to debouncing, much higher debounce constants are
tolerable, because latency does not increase with the constant, and
debounce countdowns on one key do not interfere with events on other
keys. The only negative effect of increasing the constant is that the
minimum duration of a keypress increases. Perhaps I'm just extremely
unlucky w.r.t. key switch quality, but I saw occasional bounces even
with DEBOUNCE=10; with 15, I've seen none so far. That's around 47ms,
which seems like an absolutely insane amount of time for a key to be
bouncy, but at least it works.
2017-04-27 04:29:39 +06:00
|
|
|
|
|
|
|
// Debouncing: store for each key the number of scans until it's eligible to
|
|
|
|
// change. When scanning the matrix, ignore any changes in keys that have
|
|
|
|
// already changed in the last DEBOUNCE scans.
|
|
|
|
static uint8_t debounce_matrix[MATRIX_ROWS * MATRIX_COLS];
|
2015-10-27 02:32:37 +06:00
|
|
|
|
|
|
|
static matrix_row_t read_cols(uint8_t row);
|
|
|
|
static void init_cols(void);
|
2016-05-31 19:18:40 +06:00
|
|
|
static void unselect_rows(void);
|
2015-10-27 02:32:37 +06:00
|
|
|
static void select_row(uint8_t row);
|
|
|
|
|
2018-06-13 00:27:22 +06:00
|
|
|
static uint8_t mcp23018_reset_loop;
|
|
|
|
// static uint16_t mcp23018_reset_loop;
|
2015-10-27 02:32:37 +06:00
|
|
|
|
|
|
|
#ifdef DEBUG_MATRIX_SCAN_RATE
|
|
|
|
uint32_t matrix_timer;
|
|
|
|
uint32_t matrix_scan_count;
|
|
|
|
#endif
|
|
|
|
|
|
|
|
|
2016-06-24 09:14:21 +06:00
|
|
|
__attribute__ ((weak))
|
|
|
|
void matrix_init_user(void) {}
|
|
|
|
|
|
|
|
__attribute__ ((weak))
|
|
|
|
void matrix_scan_user(void) {}
|
|
|
|
|
2015-10-27 02:32:37 +06:00
|
|
|
__attribute__ ((weak))
|
2016-03-10 22:28:34 +06:00
|
|
|
void matrix_init_kb(void) {
|
2016-06-24 09:14:21 +06:00
|
|
|
matrix_init_user();
|
2016-03-10 22:28:34 +06:00
|
|
|
}
|
2015-10-27 02:32:37 +06:00
|
|
|
|
|
|
|
__attribute__ ((weak))
|
2016-03-10 22:28:34 +06:00
|
|
|
void matrix_scan_kb(void) {
|
2016-06-24 09:14:21 +06:00
|
|
|
matrix_scan_user();
|
2016-03-10 22:28:34 +06:00
|
|
|
}
|
2015-10-27 02:32:37 +06:00
|
|
|
|
|
|
|
inline
|
|
|
|
uint8_t matrix_rows(void)
|
|
|
|
{
|
|
|
|
return MATRIX_ROWS;
|
|
|
|
}
|
|
|
|
|
|
|
|
inline
|
|
|
|
uint8_t matrix_cols(void)
|
|
|
|
{
|
|
|
|
return MATRIX_COLS;
|
|
|
|
}
|
|
|
|
|
|
|
|
void matrix_init(void)
|
|
|
|
{
|
|
|
|
// initialize row and col
|
|
|
|
|
|
|
|
mcp23018_status = init_mcp23018();
|
2015-10-27 04:54:46 +06:00
|
|
|
|
2015-10-27 02:32:37 +06:00
|
|
|
|
|
|
|
unselect_rows();
|
|
|
|
init_cols();
|
|
|
|
|
|
|
|
// initialize matrix state: all keys off
|
|
|
|
for (uint8_t i=0; i < MATRIX_ROWS; i++) {
|
|
|
|
matrix[i] = 0;
|
2018-11-20 20:55:35 +06:00
|
|
|
raw_matrix[i] = 0;
|
Change to per-key eager debouncing for ErgoDox EZ.
Empirically, waiting for N consecutive identical scans as a debouncing
strategy doesn't work very well for the ErgoDox EZ where scans are very
slow compared to most keyboards. Instead, debounce the signals by
eagerly reporting a change as soon as one scan observes it, but then
ignoring further changes from that key for the next N scans.
This is implemented by keeping an extra matrix of uint8 countdowns, such
that only keys whose countdown is currently zero are eligible to change.
When we do observe a change, we bump that key's countdown to DEBOUNCE.
During each scan, every nonzero countdown is decremented.
With this approach to debouncing, much higher debounce constants are
tolerable, because latency does not increase with the constant, and
debounce countdowns on one key do not interfere with events on other
keys. The only negative effect of increasing the constant is that the
minimum duration of a keypress increases. Perhaps I'm just extremely
unlucky w.r.t. key switch quality, but I saw occasional bounces even
with DEBOUNCE=10; with 15, I've seen none so far. That's around 47ms,
which seems like an absolutely insane amount of time for a key to be
bouncy, but at least it works.
2017-04-27 04:29:39 +06:00
|
|
|
for (uint8_t j=0; j < MATRIX_COLS; ++j) {
|
|
|
|
debounce_matrix[i * MATRIX_COLS + j] = 0;
|
|
|
|
}
|
2015-10-27 02:32:37 +06:00
|
|
|
}
|
|
|
|
|
|
|
|
#ifdef DEBUG_MATRIX_SCAN_RATE
|
|
|
|
matrix_timer = timer_read32();
|
|
|
|
matrix_scan_count = 0;
|
|
|
|
#endif
|
|
|
|
|
2016-11-14 10:02:38 +06:00
|
|
|
matrix_init_quantum();
|
2015-10-27 02:32:37 +06:00
|
|
|
|
|
|
|
}
|
|
|
|
|
2016-06-18 08:06:58 +06:00
|
|
|
void matrix_power_up(void) {
|
|
|
|
mcp23018_status = init_mcp23018();
|
|
|
|
|
|
|
|
unselect_rows();
|
|
|
|
init_cols();
|
|
|
|
|
|
|
|
// initialize matrix state: all keys off
|
|
|
|
for (uint8_t i=0; i < MATRIX_ROWS; i++) {
|
|
|
|
matrix[i] = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
#ifdef DEBUG_MATRIX_SCAN_RATE
|
|
|
|
matrix_timer = timer_read32();
|
|
|
|
matrix_scan_count = 0;
|
|
|
|
#endif
|
Change to per-key eager debouncing for ErgoDox EZ.
Empirically, waiting for N consecutive identical scans as a debouncing
strategy doesn't work very well for the ErgoDox EZ where scans are very
slow compared to most keyboards. Instead, debounce the signals by
eagerly reporting a change as soon as one scan observes it, but then
ignoring further changes from that key for the next N scans.
This is implemented by keeping an extra matrix of uint8 countdowns, such
that only keys whose countdown is currently zero are eligible to change.
When we do observe a change, we bump that key's countdown to DEBOUNCE.
During each scan, every nonzero countdown is decremented.
With this approach to debouncing, much higher debounce constants are
tolerable, because latency does not increase with the constant, and
debounce countdowns on one key do not interfere with events on other
keys. The only negative effect of increasing the constant is that the
minimum duration of a keypress increases. Perhaps I'm just extremely
unlucky w.r.t. key switch quality, but I saw occasional bounces even
with DEBOUNCE=10; with 15, I've seen none so far. That's around 47ms,
which seems like an absolutely insane amount of time for a key to be
bouncy, but at least it works.
2017-04-27 04:29:39 +06:00
|
|
|
}
|
2016-06-18 08:06:58 +06:00
|
|
|
|
Change to per-key eager debouncing for ErgoDox EZ.
Empirically, waiting for N consecutive identical scans as a debouncing
strategy doesn't work very well for the ErgoDox EZ where scans are very
slow compared to most keyboards. Instead, debounce the signals by
eagerly reporting a change as soon as one scan observes it, but then
ignoring further changes from that key for the next N scans.
This is implemented by keeping an extra matrix of uint8 countdowns, such
that only keys whose countdown is currently zero are eligible to change.
When we do observe a change, we bump that key's countdown to DEBOUNCE.
During each scan, every nonzero countdown is decremented.
With this approach to debouncing, much higher debounce constants are
tolerable, because latency does not increase with the constant, and
debounce countdowns on one key do not interfere with events on other
keys. The only negative effect of increasing the constant is that the
minimum duration of a keypress increases. Perhaps I'm just extremely
unlucky w.r.t. key switch quality, but I saw occasional bounces even
with DEBOUNCE=10; with 15, I've seen none so far. That's around 47ms,
which seems like an absolutely insane amount of time for a key to be
bouncy, but at least it works.
2017-04-27 04:29:39 +06:00
|
|
|
// Returns a matrix_row_t whose bits are set if the corresponding key should be
|
|
|
|
// eligible to change in this scan.
|
2018-11-20 20:55:35 +06:00
|
|
|
matrix_row_t debounce_mask(matrix_row_t rawcols, uint8_t row) {
|
Change to per-key eager debouncing for ErgoDox EZ.
Empirically, waiting for N consecutive identical scans as a debouncing
strategy doesn't work very well for the ErgoDox EZ where scans are very
slow compared to most keyboards. Instead, debounce the signals by
eagerly reporting a change as soon as one scan observes it, but then
ignoring further changes from that key for the next N scans.
This is implemented by keeping an extra matrix of uint8 countdowns, such
that only keys whose countdown is currently zero are eligible to change.
When we do observe a change, we bump that key's countdown to DEBOUNCE.
During each scan, every nonzero countdown is decremented.
With this approach to debouncing, much higher debounce constants are
tolerable, because latency does not increase with the constant, and
debounce countdowns on one key do not interfere with events on other
keys. The only negative effect of increasing the constant is that the
minimum duration of a keypress increases. Perhaps I'm just extremely
unlucky w.r.t. key switch quality, but I saw occasional bounces even
with DEBOUNCE=10; with 15, I've seen none so far. That's around 47ms,
which seems like an absolutely insane amount of time for a key to be
bouncy, but at least it works.
2017-04-27 04:29:39 +06:00
|
|
|
matrix_row_t result = 0;
|
2018-11-20 20:55:35 +06:00
|
|
|
matrix_row_t change = rawcols ^ raw_matrix[row];
|
|
|
|
raw_matrix[row] = rawcols;
|
|
|
|
for (uint8_t i = 0; i < MATRIX_COLS; ++i) {
|
|
|
|
if (debounce_matrix[row * MATRIX_COLS + i]) {
|
|
|
|
--debounce_matrix[row * MATRIX_COLS + i];
|
Change to per-key eager debouncing for ErgoDox EZ.
Empirically, waiting for N consecutive identical scans as a debouncing
strategy doesn't work very well for the ErgoDox EZ where scans are very
slow compared to most keyboards. Instead, debounce the signals by
eagerly reporting a change as soon as one scan observes it, but then
ignoring further changes from that key for the next N scans.
This is implemented by keeping an extra matrix of uint8 countdowns, such
that only keys whose countdown is currently zero are eligible to change.
When we do observe a change, we bump that key's countdown to DEBOUNCE.
During each scan, every nonzero countdown is decremented.
With this approach to debouncing, much higher debounce constants are
tolerable, because latency does not increase with the constant, and
debounce countdowns on one key do not interfere with events on other
keys. The only negative effect of increasing the constant is that the
minimum duration of a keypress increases. Perhaps I'm just extremely
unlucky w.r.t. key switch quality, but I saw occasional bounces even
with DEBOUNCE=10; with 15, I've seen none so far. That's around 47ms,
which seems like an absolutely insane amount of time for a key to be
bouncy, but at least it works.
2017-04-27 04:29:39 +06:00
|
|
|
} else {
|
2018-11-20 20:55:35 +06:00
|
|
|
result |= (1 << i);
|
Change to per-key eager debouncing for ErgoDox EZ.
Empirically, waiting for N consecutive identical scans as a debouncing
strategy doesn't work very well for the ErgoDox EZ where scans are very
slow compared to most keyboards. Instead, debounce the signals by
eagerly reporting a change as soon as one scan observes it, but then
ignoring further changes from that key for the next N scans.
This is implemented by keeping an extra matrix of uint8 countdowns, such
that only keys whose countdown is currently zero are eligible to change.
When we do observe a change, we bump that key's countdown to DEBOUNCE.
During each scan, every nonzero countdown is decremented.
With this approach to debouncing, much higher debounce constants are
tolerable, because latency does not increase with the constant, and
debounce countdowns on one key do not interfere with events on other
keys. The only negative effect of increasing the constant is that the
minimum duration of a keypress increases. Perhaps I'm just extremely
unlucky w.r.t. key switch quality, but I saw occasional bounces even
with DEBOUNCE=10; with 15, I've seen none so far. That's around 47ms,
which seems like an absolutely insane amount of time for a key to be
bouncy, but at least it works.
2017-04-27 04:29:39 +06:00
|
|
|
}
|
|
|
|
if (change & (1 << i)) {
|
|
|
|
debounce_matrix[row * MATRIX_COLS + i] = DEBOUNCE;
|
|
|
|
}
|
|
|
|
}
|
2018-11-20 20:55:35 +06:00
|
|
|
return result;
|
|
|
|
}
|
|
|
|
|
|
|
|
matrix_row_t debounce_read_cols(uint8_t row) {
|
|
|
|
// Read the row without debouncing filtering and store it for later usage.
|
|
|
|
matrix_row_t cols = read_cols(row);
|
|
|
|
// Get the Debounce mask.
|
|
|
|
matrix_row_t mask = debounce_mask(cols, row);
|
|
|
|
// debounce the row and return the result.
|
|
|
|
return (cols & mask) | (matrix[row] & ~mask);;
|
2016-06-18 08:06:58 +06:00
|
|
|
}
|
|
|
|
|
2015-10-27 02:32:37 +06:00
|
|
|
uint8_t matrix_scan(void)
|
|
|
|
{
|
|
|
|
if (mcp23018_status) { // if there was an error
|
2018-06-13 00:27:22 +06:00
|
|
|
if (++mcp23018_reset_loop == 0) {
|
|
|
|
// if (++mcp23018_reset_loop >= 1300) {
|
2015-10-27 02:32:37 +06:00
|
|
|
// since mcp23018_reset_loop is 8 bit - we'll try to reset once in 255 matrix scans
|
|
|
|
// this will be approx bit more frequent than once per second
|
|
|
|
print("trying to reset mcp23018\n");
|
|
|
|
mcp23018_status = init_mcp23018();
|
|
|
|
if (mcp23018_status) {
|
|
|
|
print("left side not responding\n");
|
|
|
|
} else {
|
|
|
|
print("left side attached\n");
|
|
|
|
ergodox_blink_all_leds();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
#ifdef DEBUG_MATRIX_SCAN_RATE
|
|
|
|
matrix_scan_count++;
|
|
|
|
|
|
|
|
uint32_t timer_now = timer_read32();
|
|
|
|
if (TIMER_DIFF_32(timer_now, matrix_timer)>1000) {
|
|
|
|
print("matrix scan frequency: ");
|
|
|
|
pdec(matrix_scan_count);
|
|
|
|
print("\n");
|
|
|
|
|
|
|
|
matrix_timer = timer_now;
|
|
|
|
matrix_scan_count = 0;
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
2017-10-21 05:53:59 +06:00
|
|
|
#ifdef LEFT_LEDS
|
improve ergodox ez performance
With these changes, the ergodox ez goes from 315 scans per second
when no keys are pressed (~3.17ms/scan) to 447 (~2.24ms/scan).
The changes to the pin read are just condensing the logic, and
replacing a lot of conditional operations with a single bitwise
inversion.
The change to row scanning is more significant, and merits
explanation. In general, you can only scan one row of a keyboard
at a time, because if you scan two rows, you no longer know
which row is pulling a given column down. But in the Ergodox
design, this isn't the case; the left hand is controlled by an
I2C-based GPIO expander, and the columns and rows are *completely
separate* electrically from the columns and rows on the right-hand
side.
So simply reading rows in parallel offers two significant
improvements. One is that we no longer need the 30us delay after
each right-hand row, because we're spending more than 30us
communicating with the left hand over i2c. Another is that we're
no longer wastefully sending i2c messages to the left hand
to unselect rows when no rows had actually been selected in the
first place. These delays were, between them, coming out to
nearly 30% of the time spent in each scan.
Signed-off-by: seebs <seebs@seebs.net>
2017-11-19 04:11:26 +06:00
|
|
|
mcp23018_status = ergodox_left_leds_update();
|
2017-10-21 05:53:59 +06:00
|
|
|
#endif // LEFT_LEDS
|
improve ergodox ez performance
With these changes, the ergodox ez goes from 315 scans per second
when no keys are pressed (~3.17ms/scan) to 447 (~2.24ms/scan).
The changes to the pin read are just condensing the logic, and
replacing a lot of conditional operations with a single bitwise
inversion.
The change to row scanning is more significant, and merits
explanation. In general, you can only scan one row of a keyboard
at a time, because if you scan two rows, you no longer know
which row is pulling a given column down. But in the Ergodox
design, this isn't the case; the left hand is controlled by an
I2C-based GPIO expander, and the columns and rows are *completely
separate* electrically from the columns and rows on the right-hand
side.
So simply reading rows in parallel offers two significant
improvements. One is that we no longer need the 30us delay after
each right-hand row, because we're spending more than 30us
communicating with the left hand over i2c. Another is that we're
no longer wastefully sending i2c messages to the left hand
to unselect rows when no rows had actually been selected in the
first place. These delays were, between them, coming out to
nearly 30% of the time spent in each scan.
Signed-off-by: seebs <seebs@seebs.net>
2017-11-19 04:11:26 +06:00
|
|
|
for (uint8_t i = 0; i < MATRIX_ROWS_PER_SIDE; i++) {
|
2015-10-27 02:32:37 +06:00
|
|
|
select_row(i);
|
improve ergodox ez performance
With these changes, the ergodox ez goes from 315 scans per second
when no keys are pressed (~3.17ms/scan) to 447 (~2.24ms/scan).
The changes to the pin read are just condensing the logic, and
replacing a lot of conditional operations with a single bitwise
inversion.
The change to row scanning is more significant, and merits
explanation. In general, you can only scan one row of a keyboard
at a time, because if you scan two rows, you no longer know
which row is pulling a given column down. But in the Ergodox
design, this isn't the case; the left hand is controlled by an
I2C-based GPIO expander, and the columns and rows are *completely
separate* electrically from the columns and rows on the right-hand
side.
So simply reading rows in parallel offers two significant
improvements. One is that we no longer need the 30us delay after
each right-hand row, because we're spending more than 30us
communicating with the left hand over i2c. Another is that we're
no longer wastefully sending i2c messages to the left hand
to unselect rows when no rows had actually been selected in the
first place. These delays were, between them, coming out to
nearly 30% of the time spent in each scan.
Signed-off-by: seebs <seebs@seebs.net>
2017-11-19 04:11:26 +06:00
|
|
|
// and select on left hand
|
|
|
|
select_row(i + MATRIX_ROWS_PER_SIDE);
|
|
|
|
// we don't need a 30us delay anymore, because selecting a
|
|
|
|
// left-hand row requires more than 30us for i2c.
|
2018-11-20 20:55:35 +06:00
|
|
|
|
|
|
|
// grab cols from left hand
|
|
|
|
matrix[i] = debounce_read_cols(i);
|
improve ergodox ez performance
With these changes, the ergodox ez goes from 315 scans per second
when no keys are pressed (~3.17ms/scan) to 447 (~2.24ms/scan).
The changes to the pin read are just condensing the logic, and
replacing a lot of conditional operations with a single bitwise
inversion.
The change to row scanning is more significant, and merits
explanation. In general, you can only scan one row of a keyboard
at a time, because if you scan two rows, you no longer know
which row is pulling a given column down. But in the Ergodox
design, this isn't the case; the left hand is controlled by an
I2C-based GPIO expander, and the columns and rows are *completely
separate* electrically from the columns and rows on the right-hand
side.
So simply reading rows in parallel offers two significant
improvements. One is that we no longer need the 30us delay after
each right-hand row, because we're spending more than 30us
communicating with the left hand over i2c. Another is that we're
no longer wastefully sending i2c messages to the left hand
to unselect rows when no rows had actually been selected in the
first place. These delays were, between them, coming out to
nearly 30% of the time spent in each scan.
Signed-off-by: seebs <seebs@seebs.net>
2017-11-19 04:11:26 +06:00
|
|
|
// grab cols from right hand
|
2018-11-20 20:55:35 +06:00
|
|
|
matrix[i + MATRIX_ROWS_PER_SIDE] = debounce_read_cols(i + MATRIX_ROWS_PER_SIDE);
|
|
|
|
|
Change to per-key eager debouncing for ErgoDox EZ.
Empirically, waiting for N consecutive identical scans as a debouncing
strategy doesn't work very well for the ErgoDox EZ where scans are very
slow compared to most keyboards. Instead, debounce the signals by
eagerly reporting a change as soon as one scan observes it, but then
ignoring further changes from that key for the next N scans.
This is implemented by keeping an extra matrix of uint8 countdowns, such
that only keys whose countdown is currently zero are eligible to change.
When we do observe a change, we bump that key's countdown to DEBOUNCE.
During each scan, every nonzero countdown is decremented.
With this approach to debouncing, much higher debounce constants are
tolerable, because latency does not increase with the constant, and
debounce countdowns on one key do not interfere with events on other
keys. The only negative effect of increasing the constant is that the
minimum duration of a keypress increases. Perhaps I'm just extremely
unlucky w.r.t. key switch quality, but I saw occasional bounces even
with DEBOUNCE=10; with 15, I've seen none so far. That's around 47ms,
which seems like an absolutely insane amount of time for a key to be
bouncy, but at least it works.
2017-04-27 04:29:39 +06:00
|
|
|
unselect_rows();
|
2015-10-27 02:32:37 +06:00
|
|
|
}
|
|
|
|
|
Moves features to their own files (process_*), adds tap dance feature (#460)
* non-working commit
* working
* subprojects implemented for planck
* pass a subproject variable through to c
* consolidates clueboard revisions
* thanks for letting me know about conflicts..
* turn off audio for yang's
* corrects starting paths for subprojects
* messing around with travis
* semicolon
* travis script
* travis script
* script for travis
* correct directory (probably), amend files to commit
* remove origin before adding
* git pull, correct syntax
* git checkout
* git pull origin branch
* where are we?
* where are we?
* merging
* force things to happen
* adds commit message, adds add
* rebase, no commit message
* rebase branch
* idk!
* try just pull
* fetch - merge
* specify repo branch
* checkout
* goddammit
* merge? idk
* pls
* after all
* don't split up keyboards
* syntax
* adds quick for all-keyboards
* trying out new script
* script update
* lowercase
* all keyboards
* stop replacing compiled.hex automatically
* adds if statement
* skip automated build branches
* forces push to automated build branch
* throw an add in there
* upstream?
* adds AUTOGEN
* ignore all .hex files again
* testing out new repo
* global ident
* generate script, keyboard_keymap.hex
* skip generation for now, print pandoc info, submodule update
* try trusty
* and sudo
* try generate
* updates subprojects to keyboards
* no idea
* updates to keyboards
* cleans up clueboard stuff
* setup to use local readme
* updates cluepad, planck experimental
* remove extra led.c [ci skip]
* audio and midi moved over to separate files
* chording, leader, unicode separated
* consolidate each [skip ci]
* correct include
* quantum: Add a tap dance feature (#451)
* quantum: Add a tap dance feature
With this feature one can specify keys that behave differently, based on
the amount of times they have been tapped, and when interrupted, they
get handled before the interrupter.
To make it clear how this is different from `ACTION_FUNCTION_TAP`, lets
explore a certain setup! We want one key to send `Space` on single tap,
but `Enter` on double-tap.
With `ACTION_FUNCTION_TAP`, it is quite a rain-dance to set this up, and
has the problem that when the sequence is interrupted, the interrupting
key will be send first. Thus, `SPC a` will result in `a SPC` being sent,
if they are typed within `TAPPING_TERM`. With the tap dance feature,
that'll come out as `SPC a`, correctly.
The implementation hooks into two parts of the system, to achieve this:
into `process_record_quantum()`, and the matrix scan. We need the latter
to be able to time out a tap sequence even when a key is not being
pressed, so `SPC` alone will time out and register after `TAPPING_TERM`
time.
But lets start with how to use it, first!
First, you will need `TAP_DANCE_ENABLE=yes` in your `Makefile`, because
the feature is disabled by default. This adds a little less than 1k to
the firmware size. Next, you will want to define some tap-dance keys,
which is easiest to do with the `TD()` macro, that - similar to `F()`,
takes a number, which will later be used as an index into the
`tap_dance_actions` array.
This array specifies what actions shall be taken when a tap-dance key is
in action. Currently, there are two possible options:
* `ACTION_TAP_DANCE_DOUBLE(kc1, kc2)`: Sends the `kc1` keycode when
tapped once, `kc2` otherwise.
* `ACTION_TAP_DANCE_FN(fn)`: Calls the specified function - defined in
the user keymap - with the current state of the tap-dance action.
The first option is enough for a lot of cases, that just want dual
roles. For example, `ACTION_TAP_DANCE(KC_SPC, KC_ENT)` will result in
`Space` being sent on single-tap, `Enter` otherwise.
And that's the bulk of it!
Do note, however, that this implementation does have some consequences:
keys do not register until either they reach the tapping ceiling, or
they time out. This means that if you hold the key, nothing happens, no
repeat, no nothing. It is possible to detect held state, and register an
action then too, but that's not implemented yet. Keys also unregister
immediately after being registered, so you can't even hold the second
tap. This is intentional, to be consistent.
And now, on to the explanation of how it works!
The main entry point is `process_tap_dance()`, called from
`process_record_quantum()`, which is run for every keypress, and our
handler gets to run early. This function checks whether the key pressed
is a tap-dance key. If it is not, and a tap-dance was in action, we
handle that first, and enqueue the newly pressed key. If it is a
tap-dance key, then we check if it is the same as the already active
one (if there's one active, that is). If it is not, we fire off the old
one first, then register the new one. If it was the same, we increment
the counter and the timer.
This means that you have `TAPPING_TERM` time to tap the key again, you
do not have to input all the taps within that timeframe. This allows for
longer tap counts, with minimal impact on responsiveness.
Our next stop is `matrix_scan_tap_dance()`. This handles the timeout of
tap-dance keys.
For the sake of flexibility, tap-dance actions can be either a pair of
keycodes, or a user function. The latter allows one to handle higher tap
counts, or do extra things, like blink the LEDs, fiddle with the
backlighting, and so on. This is accomplished by using an union, and
some clever macros.
In the end, lets see a full example!
```c
enum {
CT_SE = 0,
CT_CLN,
CT_EGG
};
/* Have the above three on the keymap, TD(CT_SE), etc... */
void dance_cln (qk_tap_dance_state_t *state) {
if (state->count == 1) {
register_code (KC_RSFT);
register_code (KC_SCLN);
unregister_code (KC_SCLN);
unregister_code (KC_RSFT);
} else {
register_code (KC_SCLN);
unregister_code (KC_SCLN);
reset_tap_dance (state);
}
}
void dance_egg (qk_tap_dance_state_t *state) {
if (state->count >= 100) {
SEND_STRING ("Safety dance!");
reset_tap_dance (state);
}
}
const qk_tap_dance_action_t tap_dance_actions[] = {
[CT_SE] = ACTION_TAP_DANCE_DOUBLE (KC_SPC, KC_ENT)
,[CT_CLN] = ACTION_TAP_DANCE_FN (dance_cln)
,[CT_EGG] = ACTION_TAP_DANCE_FN (dance_egg)
};
```
This addresses #426.
Signed-off-by: Gergely Nagy <algernon@madhouse-project.org>
* hhkb: Fix the build with the new tap-dance feature
Signed-off-by: Gergely Nagy <algernon@madhouse-project.org>
* tap_dance: Move process_tap_dance further down
Process the tap dance stuff after midi and audio, because those don't
process keycodes, but row/col positions.
Signed-off-by: Gergely Nagy <algernon@madhouse-project.org>
* tap_dance: Use conditionals instead of dummy functions
To be consistent with how the rest of the quantum features are
implemented, use ifdefs instead of dummy functions.
Signed-off-by: Gergely Nagy <algernon@madhouse-project.org>
* Merge branch 'master' into quantum-keypress-process
# Conflicts:
# Makefile
# keyboards/planck/rev3/config.h
# keyboards/planck/rev4/config.h
* update build script
2016-06-30 03:49:41 +06:00
|
|
|
matrix_scan_quantum();
|
2015-10-27 02:32:37 +06:00
|
|
|
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
Change to per-key eager debouncing for ErgoDox EZ.
Empirically, waiting for N consecutive identical scans as a debouncing
strategy doesn't work very well for the ErgoDox EZ where scans are very
slow compared to most keyboards. Instead, debounce the signals by
eagerly reporting a change as soon as one scan observes it, but then
ignoring further changes from that key for the next N scans.
This is implemented by keeping an extra matrix of uint8 countdowns, such
that only keys whose countdown is currently zero are eligible to change.
When we do observe a change, we bump that key's countdown to DEBOUNCE.
During each scan, every nonzero countdown is decremented.
With this approach to debouncing, much higher debounce constants are
tolerable, because latency does not increase with the constant, and
debounce countdowns on one key do not interfere with events on other
keys. The only negative effect of increasing the constant is that the
minimum duration of a keypress increases. Perhaps I'm just extremely
unlucky w.r.t. key switch quality, but I saw occasional bounces even
with DEBOUNCE=10; with 15, I've seen none so far. That's around 47ms,
which seems like an absolutely insane amount of time for a key to be
bouncy, but at least it works.
2017-04-27 04:29:39 +06:00
|
|
|
bool matrix_is_modified(void) // deprecated and evidently not called.
|
2015-10-27 02:32:37 +06:00
|
|
|
{
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
inline
|
|
|
|
bool matrix_is_on(uint8_t row, uint8_t col)
|
|
|
|
{
|
|
|
|
return (matrix[row] & ((matrix_row_t)1<<col));
|
|
|
|
}
|
|
|
|
|
|
|
|
inline
|
|
|
|
matrix_row_t matrix_get_row(uint8_t row)
|
|
|
|
{
|
|
|
|
return matrix[row];
|
|
|
|
}
|
|
|
|
|
|
|
|
void matrix_print(void)
|
|
|
|
{
|
|
|
|
print("\nr/c 0123456789ABCDEF\n");
|
|
|
|
for (uint8_t row = 0; row < MATRIX_ROWS; row++) {
|
|
|
|
phex(row); print(": ");
|
|
|
|
pbin_reverse16(matrix_get_row(row));
|
|
|
|
print("\n");
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
uint8_t matrix_key_count(void)
|
|
|
|
{
|
|
|
|
uint8_t count = 0;
|
|
|
|
for (uint8_t i = 0; i < MATRIX_ROWS; i++) {
|
|
|
|
count += bitpop16(matrix[i]);
|
|
|
|
}
|
|
|
|
return count;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Column pin configuration
|
|
|
|
*
|
|
|
|
* Teensy
|
|
|
|
* col: 0 1 2 3 4 5
|
2015-10-27 04:54:46 +06:00
|
|
|
* pin: F0 F1 F4 F5 F6 F7
|
2015-10-27 02:32:37 +06:00
|
|
|
*
|
|
|
|
* MCP23018
|
|
|
|
* col: 0 1 2 3 4 5
|
2015-10-27 04:54:46 +06:00
|
|
|
* pin: B5 B4 B3 B2 B1 B0
|
2015-10-27 02:32:37 +06:00
|
|
|
*/
|
|
|
|
static void init_cols(void)
|
|
|
|
{
|
|
|
|
// init on mcp23018
|
|
|
|
// not needed, already done as part of init_mcp23018()
|
|
|
|
|
|
|
|
// init on teensy
|
|
|
|
// Input with pull-up(DDR:0, PORT:1)
|
|
|
|
DDRF &= ~(1<<7 | 1<<6 | 1<<5 | 1<<4 | 1<<1 | 1<<0);
|
|
|
|
PORTF |= (1<<7 | 1<<6 | 1<<5 | 1<<4 | 1<<1 | 1<<0);
|
|
|
|
}
|
|
|
|
|
|
|
|
static matrix_row_t read_cols(uint8_t row)
|
|
|
|
{
|
|
|
|
if (row < 7) {
|
|
|
|
if (mcp23018_status) { // if there was an error
|
|
|
|
return 0;
|
|
|
|
} else {
|
|
|
|
uint8_t data = 0;
|
2018-06-24 00:18:47 +06:00
|
|
|
mcp23018_status = i2c_start(I2C_ADDR_WRITE, ERGODOX_EZ_I2C_TIMEOUT); if (mcp23018_status) goto out;
|
|
|
|
mcp23018_status = i2c_write(GPIOB, ERGODOX_EZ_I2C_TIMEOUT); if (mcp23018_status) goto out;
|
|
|
|
mcp23018_status = i2c_start(I2C_ADDR_READ, ERGODOX_EZ_I2C_TIMEOUT); if (mcp23018_status) goto out;
|
|
|
|
mcp23018_status = i2c_read_nack(ERGODOX_EZ_I2C_TIMEOUT); if (mcp23018_status < 0) goto out;
|
2018-06-23 07:26:30 +06:00
|
|
|
data = ~((uint8_t)mcp23018_status);
|
2018-06-24 00:18:47 +06:00
|
|
|
mcp23018_status = I2C_STATUS_SUCCESS;
|
2015-10-27 02:32:37 +06:00
|
|
|
out:
|
2018-06-24 00:18:47 +06:00
|
|
|
i2c_stop(ERGODOX_EZ_I2C_TIMEOUT);
|
2015-10-27 02:32:37 +06:00
|
|
|
return data;
|
|
|
|
}
|
|
|
|
} else {
|
improve ergodox ez performance
With these changes, the ergodox ez goes from 315 scans per second
when no keys are pressed (~3.17ms/scan) to 447 (~2.24ms/scan).
The changes to the pin read are just condensing the logic, and
replacing a lot of conditional operations with a single bitwise
inversion.
The change to row scanning is more significant, and merits
explanation. In general, you can only scan one row of a keyboard
at a time, because if you scan two rows, you no longer know
which row is pulling a given column down. But in the Ergodox
design, this isn't the case; the left hand is controlled by an
I2C-based GPIO expander, and the columns and rows are *completely
separate* electrically from the columns and rows on the right-hand
side.
So simply reading rows in parallel offers two significant
improvements. One is that we no longer need the 30us delay after
each right-hand row, because we're spending more than 30us
communicating with the left hand over i2c. Another is that we're
no longer wastefully sending i2c messages to the left hand
to unselect rows when no rows had actually been selected in the
first place. These delays were, between them, coming out to
nearly 30% of the time spent in each scan.
Signed-off-by: seebs <seebs@seebs.net>
2017-11-19 04:11:26 +06:00
|
|
|
/* read from teensy
|
|
|
|
* bitmask is 0b11110011, but we want those all
|
|
|
|
* in the lower six bits.
|
|
|
|
* we'll return 1s for the top two, but that's harmless.
|
|
|
|
*/
|
|
|
|
|
|
|
|
return ~((PINF & 0x03) | ((PINF & 0xF0) >> 2));
|
2015-10-27 02:32:37 +06:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Row pin configuration
|
|
|
|
*
|
|
|
|
* Teensy
|
|
|
|
* row: 7 8 9 10 11 12 13
|
|
|
|
* pin: B0 B1 B2 B3 D2 D3 C6
|
|
|
|
*
|
|
|
|
* MCP23018
|
|
|
|
* row: 0 1 2 3 4 5 6
|
|
|
|
* pin: A0 A1 A2 A3 A4 A5 A6
|
|
|
|
*/
|
|
|
|
static void unselect_rows(void)
|
|
|
|
{
|
Don't "unselect" left-hand rows
"unselecting" left-hand rows is a wasted i2c transaction.
On the left-hand side, the ergodox uses a GPIO expander. It
does *not* change "direction" (input/output) of pins, it just
sets pins high or low.
But all the pins are written at once. There's no way to
change just one pin's value; you send a full byte of all eight
row pins. (Not all of them are in use, but that doesn't matter.)
So every pin is either +V or ground. This is in contrast
with the right-hand side, which is using input mode to make pins
be neutral.
So there's no need to "deselect" the rows on the left side
at all. To select row 0, you set the GPIO register for the
rows to 0xFE. The previous code would then set it back to
0xFF, then set it to 0xFD on the next cycle. But we can just
omit the intervening step, and set it to 0xFD next cycle,
and get the same results.
And yes, I tested that the keyboard still works.
On my system, scan rate as reported by DEBUG_SCAN_RATE goes
from 445 or so to 579 or so, thus, from ~2.24ms to ~1.73ms.
Signed-off-by: seebs <seebs@seebs.net>
2017-12-05 22:55:05 +06:00
|
|
|
// no need to unselect on mcp23018, because the select step sets all
|
|
|
|
// the other row bits high, and it's not changing to a different
|
|
|
|
// direction
|
2015-10-27 02:32:37 +06:00
|
|
|
|
|
|
|
// unselect on teensy
|
|
|
|
// Hi-Z(DDR:0, PORT:0) to unselect
|
|
|
|
DDRB &= ~(1<<0 | 1<<1 | 1<<2 | 1<<3);
|
|
|
|
PORTB &= ~(1<<0 | 1<<1 | 1<<2 | 1<<3);
|
|
|
|
DDRD &= ~(1<<2 | 1<<3);
|
|
|
|
PORTD &= ~(1<<2 | 1<<3);
|
|
|
|
DDRC &= ~(1<<6);
|
|
|
|
PORTC &= ~(1<<6);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void select_row(uint8_t row)
|
|
|
|
{
|
|
|
|
if (row < 7) {
|
|
|
|
// select on mcp23018
|
|
|
|
if (mcp23018_status) { // if there was an error
|
|
|
|
// do nothing
|
|
|
|
} else {
|
|
|
|
// set active row low : 0
|
|
|
|
// set other rows hi-Z : 1
|
2018-06-24 00:18:47 +06:00
|
|
|
mcp23018_status = i2c_start(I2C_ADDR_WRITE, ERGODOX_EZ_I2C_TIMEOUT); if (mcp23018_status) goto out;
|
|
|
|
mcp23018_status = i2c_write(GPIOA, ERGODOX_EZ_I2C_TIMEOUT); if (mcp23018_status) goto out;
|
|
|
|
mcp23018_status = i2c_write(0xFF & ~(1<<row), ERGODOX_EZ_I2C_TIMEOUT); if (mcp23018_status) goto out;
|
2015-10-27 02:32:37 +06:00
|
|
|
out:
|
2018-06-24 00:18:47 +06:00
|
|
|
i2c_stop(ERGODOX_EZ_I2C_TIMEOUT);
|
2015-10-27 02:32:37 +06:00
|
|
|
}
|
|
|
|
} else {
|
|
|
|
// select on teensy
|
|
|
|
// Output low(DDR:1, PORT:0) to select
|
|
|
|
switch (row) {
|
|
|
|
case 7:
|
|
|
|
DDRB |= (1<<0);
|
|
|
|
PORTB &= ~(1<<0);
|
|
|
|
break;
|
|
|
|
case 8:
|
|
|
|
DDRB |= (1<<1);
|
|
|
|
PORTB &= ~(1<<1);
|
|
|
|
break;
|
|
|
|
case 9:
|
|
|
|
DDRB |= (1<<2);
|
|
|
|
PORTB &= ~(1<<2);
|
|
|
|
break;
|
|
|
|
case 10:
|
|
|
|
DDRB |= (1<<3);
|
|
|
|
PORTB &= ~(1<<3);
|
|
|
|
break;
|
|
|
|
case 11:
|
|
|
|
DDRD |= (1<<2);
|
2018-11-21 21:49:28 +06:00
|
|
|
PORTD &= ~(1<<2);
|
2015-10-27 02:32:37 +06:00
|
|
|
break;
|
|
|
|
case 12:
|
|
|
|
DDRD |= (1<<3);
|
|
|
|
PORTD &= ~(1<<3);
|
|
|
|
break;
|
|
|
|
case 13:
|
|
|
|
DDRC |= (1<<6);
|
|
|
|
PORTC &= ~(1<<6);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|