From ecc36d8dc7ccb2dcdedcc89bedb6898e2f15f0de Mon Sep 17 00:00:00 2001
From: mtei <2170248+mtei@users.noreply.github.com>
Date: Sat, 27 Jun 2020 20:27:08 +0900
Subject: [PATCH 01/17] fix matrix_io_delay() timing in quantum/matrix.c

---
 quantum/matrix.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/quantum/matrix.c b/quantum/matrix.c
index cab0d2ddcafa..807dd68f8e9a 100644
--- a/quantum/matrix.c
+++ b/quantum/matrix.c
@@ -109,7 +109,6 @@ static bool read_cols_on_row(matrix_row_t current_matrix[], uint8_t current_row)
 
     // Select row and wait for row selecton to stabilize
     select_row(current_row);
-    matrix_io_delay();
 
     // For each col...
     for (uint8_t col_index = 0; col_index < MATRIX_COLS; col_index++) {
@@ -122,6 +121,9 @@ static bool read_cols_on_row(matrix_row_t current_matrix[], uint8_t current_row)
 
     // Unselect row
     unselect_row(current_row);
+    if( current_row + 1 < MATRIX_ROWS ) {
+        matrix_io_delay();
+    }
 
     // If the row has changed, store the row and return the changed flag.
     if (current_matrix[current_row] != current_row_value) {
@@ -159,7 +161,6 @@ static bool read_rows_on_col(matrix_row_t current_matrix[], uint8_t current_col)
 
     // Select col and wait for col selecton to stabilize
     select_col(current_col);
-    matrix_io_delay();
 
     // For each row...
     for (uint8_t row_index = 0; row_index < MATRIX_ROWS; row_index++) {
@@ -185,6 +186,9 @@ static bool read_rows_on_col(matrix_row_t current_matrix[], uint8_t current_col)
 
     // Unselect col
     unselect_col(current_col);
+    if( current_col + 1 < MATRIX_COLS ) {
+        matrix_io_delay();
+    }
 
     return matrix_changed;
 }

From 45fe92cac17e9e9f2cef28b98af99a58f553e989 Mon Sep 17 00:00:00 2001
From: mtei <2170248+mtei@users.noreply.github.com>
Date: Thu, 2 Jul 2020 13:24:23 +0900
Subject: [PATCH 02/17] Updated comments explaining the need for
 matrix_io_delay() in quantum/matrix.c

---
 quantum/matrix.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/quantum/matrix.c b/quantum/matrix.c
index 807dd68f8e9a..662ae20269cc 100644
--- a/quantum/matrix.c
+++ b/quantum/matrix.c
@@ -107,7 +107,7 @@ static bool read_cols_on_row(matrix_row_t current_matrix[], uint8_t current_row)
     // Start with a clear matrix row
     matrix_row_t current_row_value = 0;
 
-    // Select row and wait for row selecton to stabilize
+    // Select row
     select_row(current_row);
 
     // For each col...
@@ -122,7 +122,7 @@ static bool read_cols_on_row(matrix_row_t current_matrix[], uint8_t current_row)
     // Unselect row
     unselect_row(current_row);
     if( current_row + 1 < MATRIX_ROWS ) {
-        matrix_io_delay();
+        matrix_io_delay(); // wait for row signal to HIGH
     }
 
     // If the row has changed, store the row and return the changed flag.
@@ -159,7 +159,7 @@ static void init_pins(void) {
 static bool read_rows_on_col(matrix_row_t current_matrix[], uint8_t current_col) {
     bool matrix_changed = false;
 
-    // Select col and wait for col selecton to stabilize
+    // Select col
     select_col(current_col);
 
     // For each row...
@@ -187,7 +187,7 @@ static bool read_rows_on_col(matrix_row_t current_matrix[], uint8_t current_col)
     // Unselect col
     unselect_col(current_col);
     if( current_col + 1 < MATRIX_COLS ) {
-        matrix_io_delay();
+        matrix_io_delay(); // wait for col signal to HIGH
     }
 
     return matrix_changed;

From b806d574901721ff73329430a0a8235dea30e394 Mon Sep 17 00:00:00 2001
From: mtei <2170248+mtei@users.noreply.github.com>
Date: Thu, 2 Jul 2020 13:50:24 +0900
Subject: [PATCH 03/17] fix matrix_io_delay() timing in
 quantum/split_common/matrix.c

---
 quantum/split_common/matrix.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)

diff --git a/quantum/split_common/matrix.c b/quantum/split_common/matrix.c
index cd5a024c3df4..0611482d7fa4 100644
--- a/quantum/split_common/matrix.c
+++ b/quantum/split_common/matrix.c
@@ -120,9 +120,8 @@ static bool read_cols_on_row(matrix_row_t current_matrix[], uint8_t current_row)
     // Start with a clear matrix row
     matrix_row_t current_row_value = 0;
 
-    // Select row and wait for row selecton to stabilize
+    // Select row
     select_row(current_row);
-    matrix_io_delay();
 
     // For each col...
     for (uint8_t col_index = 0; col_index < MATRIX_COLS; col_index++) {
@@ -135,6 +134,9 @@ static bool read_cols_on_row(matrix_row_t current_matrix[], uint8_t current_row)
 
     // Unselect row
     unselect_row(current_row);
+    if( current_row + 1 < MATRIX_ROWS ) {
+        matrix_io_delay(); // wait for row signal to HIGH
+    }
 
     // If the row has changed, store the row and return the changed flag.
     if (current_matrix[current_row] != current_row_value) {
@@ -170,9 +172,8 @@ static void init_pins(void) {
 static bool read_rows_on_col(matrix_row_t current_matrix[], uint8_t current_col) {
     bool matrix_changed = false;
 
-    // Select col and wait for col selecton to stabilize
+    // Select col
     select_col(current_col);
-    matrix_io_delay();
 
     // For each row...
     for (uint8_t row_index = 0; row_index < ROWS_PER_HAND; row_index++) {
@@ -198,6 +199,9 @@ static bool read_rows_on_col(matrix_row_t current_matrix[], uint8_t current_col)
 
     // Unselect col
     unselect_col(current_col);
+    if( current_col + 1 < MATRIX_COLS ) {
+        matrix_io_delay(); // wait for col signal to HIGH
+    }
 
     return matrix_changed;
 }

From 0f28e2c3699499af46b820d0b2d9f9998bad884a Mon Sep 17 00:00:00 2001
From: Takeshi ISHII <2170248+mtei@users.noreply.github.com>
Date: Thu, 2 Jul 2020 18:19:24 +0900
Subject: [PATCH 04/17] Update quantum/matrix.c

Co-authored-by: Ryan <fauxpark@gmail.com>
---
 quantum/matrix.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/quantum/matrix.c b/quantum/matrix.c
index 662ae20269cc..e12ea753697b 100644
--- a/quantum/matrix.c
+++ b/quantum/matrix.c
@@ -121,8 +121,8 @@ static bool read_cols_on_row(matrix_row_t current_matrix[], uint8_t current_row)
 
     // Unselect row
     unselect_row(current_row);
-    if( current_row + 1 < MATRIX_ROWS ) {
-        matrix_io_delay(); // wait for row signal to HIGH
+    if (current_row + 1 < MATRIX_ROWS) {
+        matrix_io_delay();  // wait for row signal to go HIGH
     }
 
     // If the row has changed, store the row and return the changed flag.

From ab50411fae620dd4c9310e5bfbe413558ae8f7ce Mon Sep 17 00:00:00 2001
From: Takeshi ISHII <2170248+mtei@users.noreply.github.com>
Date: Thu, 2 Jul 2020 18:19:40 +0900
Subject: [PATCH 05/17] Update quantum/split_common/matrix.c

Co-authored-by: Ryan <fauxpark@gmail.com>
---
 quantum/split_common/matrix.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/quantum/split_common/matrix.c b/quantum/split_common/matrix.c
index 0611482d7fa4..76b3bd3a59b0 100644
--- a/quantum/split_common/matrix.c
+++ b/quantum/split_common/matrix.c
@@ -199,8 +199,8 @@ static bool read_rows_on_col(matrix_row_t current_matrix[], uint8_t current_col)
 
     // Unselect col
     unselect_col(current_col);
-    if( current_col + 1 < MATRIX_COLS ) {
-        matrix_io_delay(); // wait for col signal to HIGH
+    if (current_col + 1 < MATRIX_COLS) {
+        matrix_io_delay();  // wait for col signal to go HIGH
     }
 
     return matrix_changed;

From b8093393f8d646079b1a794c32f9739283420845 Mon Sep 17 00:00:00 2001
From: Takeshi ISHII <2170248+mtei@users.noreply.github.com>
Date: Thu, 2 Jul 2020 18:20:05 +0900
Subject: [PATCH 06/17] Update quantum/matrix.c

Co-authored-by: Ryan <fauxpark@gmail.com>
---
 quantum/matrix.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/quantum/matrix.c b/quantum/matrix.c
index e12ea753697b..697b63a729dc 100644
--- a/quantum/matrix.c
+++ b/quantum/matrix.c
@@ -186,8 +186,8 @@ static bool read_rows_on_col(matrix_row_t current_matrix[], uint8_t current_col)
 
     // Unselect col
     unselect_col(current_col);
-    if( current_col + 1 < MATRIX_COLS ) {
-        matrix_io_delay(); // wait for col signal to HIGH
+    if (current_col + 1 < MATRIX_COLS) {
+        matrix_io_delay();  // wait for col signal to go HIGH
     }
 
     return matrix_changed;

From 0614cefc1596b4da49d1ebcff19dad582643b2f7 Mon Sep 17 00:00:00 2001
From: Takeshi ISHII <2170248+mtei@users.noreply.github.com>
Date: Thu, 2 Jul 2020 18:20:18 +0900
Subject: [PATCH 07/17] Update quantum/split_common/matrix.c

Co-authored-by: Ryan <fauxpark@gmail.com>
---
 quantum/split_common/matrix.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/quantum/split_common/matrix.c b/quantum/split_common/matrix.c
index 76b3bd3a59b0..0ea53030b90b 100644
--- a/quantum/split_common/matrix.c
+++ b/quantum/split_common/matrix.c
@@ -134,8 +134,8 @@ static bool read_cols_on_row(matrix_row_t current_matrix[], uint8_t current_row)
 
     // Unselect row
     unselect_row(current_row);
-    if( current_row + 1 < MATRIX_ROWS ) {
-        matrix_io_delay(); // wait for row signal to HIGH
+    if (current_row + 1 < MATRIX_ROWS) {
+        matrix_io_delay();  // wait for row signal to go HIGH
     }
 
     // If the row has changed, store the row and return the changed flag.

From c5ec5b718a53ff96122cadadcfc539e5287308d1 Mon Sep 17 00:00:00 2001
From: mtei <2170248+mtei@users.noreply.github.com>
Date: Sat, 14 Nov 2020 12:00:13 +0900
Subject: [PATCH 08/17] add waitOutputPinValid() and wait_cpuclock() into
 quantum/quantum.h and tmk_core/common/wait.h

---
 quantum/quantum.h      |  7 ++++++
 tmk_core/common/wait.h | 52 ++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 59 insertions(+)

diff --git a/quantum/quantum.h b/quantum/quantum.h
index cb0af306ac93..e70c4dd28c78 100644
--- a/quantum/quantum.h
+++ b/quantum/quantum.h
@@ -208,6 +208,8 @@ typedef uint8_t pin_t;
 
 #    define togglePin(pin) (PORTx_ADDRESS(pin) ^= _BV((pin)&0xF))
 
+#    define waitOutputPinValid() wait_cpuclock(1)
+
 #elif defined(PROTOCOL_CHIBIOS)
 typedef ioline_t pin_t;
 
@@ -223,6 +225,11 @@ typedef ioline_t pin_t;
 #    define readPin(pin) palReadLine(pin)
 
 #    define togglePin(pin) palToggleLine(pin)
+
+#endif
+
+#if defined(__ARMEL__) || defined(__ARMEB__)
+#    define waitOutputPinValid() wait_cpuclock(STM32_SYSCLK/1000000L / 4)
 #endif
 
 // Atomic macro to help make GPIO and other controls atomic.
diff --git a/tmk_core/common/wait.h b/tmk_core/common/wait.h
index c82cd2d65a29..26bd0675680d 100644
--- a/tmk_core/common/wait.h
+++ b/tmk_core/common/wait.h
@@ -7,10 +7,62 @@
 extern "C" {
 #endif
 
+#if defined(__ARMEL__) || defined(__ARMEB__)
+#    ifndef __OPTIMIZE__
+#        error "Compiler optimizations disabled; wait_cpuclock() won't work as designed"
+#    endif
+
+#    define CLOCK_DELAY_NOP8 "nop\n\t nop\n\t nop\n\t nop\n\t   nop\n\t nop\n\t nop\n\t nop\n\t"
+
+#define WAIT_CPUCLOCK_DEFINED
+
+__attribute__((always_inline))
+static inline void wait_cpuclock(unsigned int n) { /* n: 1..135 */
+    /* The argument n must be a constant expression.
+     * That way, compiler optimization will remove unnecessary code. */
+    if (n < 1) { return; }
+    if (n > 8) {
+        unsigned int n8 = n/8;
+        n = n - n8*8;
+        switch (n8) {
+        case 16: asm volatile (CLOCK_DELAY_NOP8::: "memory");
+        case 15: asm volatile (CLOCK_DELAY_NOP8::: "memory");
+        case 14: asm volatile (CLOCK_DELAY_NOP8::: "memory");
+        case 13: asm volatile (CLOCK_DELAY_NOP8::: "memory");
+        case 12: asm volatile (CLOCK_DELAY_NOP8::: "memory");
+        case 11: asm volatile (CLOCK_DELAY_NOP8::: "memory");
+        case 10: asm volatile (CLOCK_DELAY_NOP8::: "memory");
+        case  9: asm volatile (CLOCK_DELAY_NOP8::: "memory");
+        case  8: asm volatile (CLOCK_DELAY_NOP8::: "memory");
+        case  7: asm volatile (CLOCK_DELAY_NOP8::: "memory");
+        case  6: asm volatile (CLOCK_DELAY_NOP8::: "memory");
+        case  5: asm volatile (CLOCK_DELAY_NOP8::: "memory");
+        case  4: asm volatile (CLOCK_DELAY_NOP8::: "memory");
+        case  3: asm volatile (CLOCK_DELAY_NOP8::: "memory");
+        case  2: asm volatile (CLOCK_DELAY_NOP8::: "memory");
+        case  1: asm volatile (CLOCK_DELAY_NOP8::: "memory");
+        case  0: break;
+        }
+    }
+    switch (n) {
+    case 8: asm volatile ("nop"::: "memory");
+    case 7: asm volatile ("nop"::: "memory");
+    case 6: asm volatile ("nop"::: "memory");
+    case 5: asm volatile ("nop"::: "memory");
+    case 4: asm volatile ("nop"::: "memory");
+    case 3: asm volatile ("nop"::: "memory");
+    case 2: asm volatile ("nop"::: "memory");
+    case 1: asm volatile ("nop"::: "memory");
+    case 0: break;
+    }
+}
+#endif
+
 #if defined(__AVR__)
 #    include <util/delay.h>
 #    define wait_ms(ms) _delay_ms(ms)
 #    define wait_us(us) _delay_us(us)
+#    define wait_cpuclock(x) __builtin_avr_delay_cycles(x)
 #elif defined PROTOCOL_CHIBIOS
 #    include "ch.h"
 #    define wait_ms(ms)                     \

From dfb4c54a6d56103492cfbc1d930b6decb60f9cf2 Mon Sep 17 00:00:00 2001
From: mtei <2170248+mtei@users.noreply.github.com>
Date: Sat, 14 Nov 2020 12:36:29 +0900
Subject: [PATCH 09/17] add matrix_output_select_delay() and
 matrix_output_unselect_delay()

---
 quantum/matrix.c              | 6 ++++--
 quantum/matrix_common.c       | 4 ++++
 quantum/split_common/matrix.c | 6 ++++--
 3 files changed, 12 insertions(+), 4 deletions(-)

diff --git a/quantum/matrix.c b/quantum/matrix.c
index 697b63a729dc..8778fb036477 100644
--- a/quantum/matrix.c
+++ b/quantum/matrix.c
@@ -109,6 +109,7 @@ static bool read_cols_on_row(matrix_row_t current_matrix[], uint8_t current_row)
 
     // Select row
     select_row(current_row);
+    matrix_output_select_delay();
 
     // For each col...
     for (uint8_t col_index = 0; col_index < MATRIX_COLS; col_index++) {
@@ -122,7 +123,7 @@ static bool read_cols_on_row(matrix_row_t current_matrix[], uint8_t current_row)
     // Unselect row
     unselect_row(current_row);
     if (current_row + 1 < MATRIX_ROWS) {
-        matrix_io_delay();  // wait for row signal to go HIGH
+        matrix_output_unselect_delay();  // wait for row signal to go HIGH
     }
 
     // If the row has changed, store the row and return the changed flag.
@@ -161,6 +162,7 @@ static bool read_rows_on_col(matrix_row_t current_matrix[], uint8_t current_col)
 
     // Select col
     select_col(current_col);
+    matrix_output_select_delay();
 
     // For each row...
     for (uint8_t row_index = 0; row_index < MATRIX_ROWS; row_index++) {
@@ -187,7 +189,7 @@ static bool read_rows_on_col(matrix_row_t current_matrix[], uint8_t current_col)
     // Unselect col
     unselect_col(current_col);
     if (current_col + 1 < MATRIX_COLS) {
-        matrix_io_delay();  // wait for col signal to go HIGH
+        matrix_output_unselect_delay();  // wait for row signal to go HIGH
     }
 
     return matrix_changed;
diff --git a/quantum/matrix_common.c b/quantum/matrix_common.c
index 15f1e0e82e76..d2909ec6d3b6 100644
--- a/quantum/matrix_common.c
+++ b/quantum/matrix_common.c
@@ -83,8 +83,12 @@ uint8_t matrix_key_count(void) {
     return count;
 }
 
+/*　`matrix_io_delay ()` exists for backwards compatibility. From now on, use matrix_output_unselect_delay().　*/
 __attribute__((weak)) void matrix_io_delay(void) { wait_us(MATRIX_IO_DELAY); }
 
+__attribute__((weak)) void matrix_output_select_delay(void) { waitOutputPinValid(); }
+__attribute__((weak)) void matrix_output_unselect_delay(void) { matrix_io_delay(); }
+
 // CUSTOM MATRIX 'LITE'
 __attribute__((weak)) void matrix_init_custom(void) {}
 
diff --git a/quantum/split_common/matrix.c b/quantum/split_common/matrix.c
index 0ea53030b90b..5d2c4caba11e 100644
--- a/quantum/split_common/matrix.c
+++ b/quantum/split_common/matrix.c
@@ -122,6 +122,7 @@ static bool read_cols_on_row(matrix_row_t current_matrix[], uint8_t current_row)
 
     // Select row
     select_row(current_row);
+    matrix_output_select_delay();
 
     // For each col...
     for (uint8_t col_index = 0; col_index < MATRIX_COLS; col_index++) {
@@ -135,7 +136,7 @@ static bool read_cols_on_row(matrix_row_t current_matrix[], uint8_t current_row)
     // Unselect row
     unselect_row(current_row);
     if (current_row + 1 < MATRIX_ROWS) {
-        matrix_io_delay();  // wait for row signal to go HIGH
+        matrix_output_unselect_delay();  // wait for row signal to go HIGH
     }
 
     // If the row has changed, store the row and return the changed flag.
@@ -174,6 +175,7 @@ static bool read_rows_on_col(matrix_row_t current_matrix[], uint8_t current_col)
 
     // Select col
     select_col(current_col);
+    matrix_output_select_delay();
 
     // For each row...
     for (uint8_t row_index = 0; row_index < ROWS_PER_HAND; row_index++) {
@@ -200,7 +202,7 @@ static bool read_rows_on_col(matrix_row_t current_matrix[], uint8_t current_col)
     // Unselect col
     unselect_col(current_col);
     if (current_col + 1 < MATRIX_COLS) {
-        matrix_io_delay();  // wait for col signal to go HIGH
+        matrix_output_unselect_delay();  // wait for row signal to go HIGH
     }
 
     return matrix_changed;

From 77a2cf3cf83d42dc8c4aed2b4e0671e228568563 Mon Sep 17 00:00:00 2001
From: mtei <2170248+mtei@users.noreply.github.com>
Date: Sat, 14 Nov 2020 13:10:33 +0900
Subject: [PATCH 10/17] fix quantum/matrix_common.c, tmk_core/common/matrix.h

---
 quantum/matrix_common.c  | 1 +
 tmk_core/common/matrix.h | 3 +++
 2 files changed, 4 insertions(+)

diff --git a/quantum/matrix_common.c b/quantum/matrix_common.c
index d2909ec6d3b6..d5a857084464 100644
--- a/quantum/matrix_common.c
+++ b/quantum/matrix_common.c
@@ -1,3 +1,4 @@
+#include "quantum.h"
 #include "matrix.h"
 #include "debounce.h"
 #include "wait.h"
diff --git a/tmk_core/common/matrix.h b/tmk_core/common/matrix.h
index 31ec844302c8..489333f60e81 100644
--- a/tmk_core/common/matrix.h
+++ b/tmk_core/common/matrix.h
@@ -57,6 +57,9 @@ matrix_row_t matrix_get_row(uint8_t row);
 /* print matrix for debug */
 void matrix_print(void);
 /* delay between changing matrix pin state and reading values */
+void matrix_output_select_delay(void);
+void matrix_output_unselect_delay(void);
+/* only for backwards compatibility. delay between changing matrix pin state and reading values */
 void matrix_io_delay(void);
 
 /* power control */

From e7fc73b4b787a3ed1ce17327c26c5da9afc0d1e2 Mon Sep 17 00:00:00 2001
From: mtei <2170248+mtei@users.noreply.github.com>
Date: Sat, 14 Nov 2020 13:49:21 +0900
Subject: [PATCH 11/17] fix tmk_core/common/wait.h

---
 tmk_core/common/wait.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/tmk_core/common/wait.h b/tmk_core/common/wait.h
index 26bd0675680d..7d0c0e3b70a4 100644
--- a/tmk_core/common/wait.h
+++ b/tmk_core/common/wait.h
@@ -9,7 +9,7 @@ extern "C" {
 
 #if defined(__ARMEL__) || defined(__ARMEB__)
 #    ifndef __OPTIMIZE__
-#        error "Compiler optimizations disabled; wait_cpuclock() won't work as designed"
+#        warning "Compiler optimizations disabled; wait_cpuclock() won't work as designed"
 #    endif
 
 #    define CLOCK_DELAY_NOP8 "nop\n\t nop\n\t nop\n\t nop\n\t   nop\n\t nop\n\t nop\n\t nop\n\t"

From c8ef016d260a313f11cc77850c840c49e0e0aa3d Mon Sep 17 00:00:00 2001
From: mtei <2170248+mtei@users.noreply.github.com>
Date: Sat, 14 Nov 2020 18:04:49 +0900
Subject: [PATCH 12/17] fix quantum/quantum.h, tmk_core/common/wait.h

---
 quantum/quantum.h      | 7 ++++++-
 tmk_core/common/wait.h | 2 +-
 2 files changed, 7 insertions(+), 2 deletions(-)

diff --git a/quantum/quantum.h b/quantum/quantum.h
index e70c4dd28c78..7f13cc7cfa9e 100644
--- a/quantum/quantum.h
+++ b/quantum/quantum.h
@@ -229,7 +229,12 @@ typedef ioline_t pin_t;
 #endif
 
 #if defined(__ARMEL__) || defined(__ARMEB__)
-#    define waitOutputPinValid() wait_cpuclock(STM32_SYSCLK/1000000L / 4)
+#    if  defined(STM32_SYSCLK)
+#        define WAITOUTPUTPINVALID_DELAY (STM32_SYSCLK/1000000L / 4)
+#    elif defined(KINETIS_SYSCLK_FREQUENCY)
+#        define WAITOUTPUTPINVALID_DELAY (KINETIS_SYSCLK_FREQUENCY/1000000L / 4)
+#    endif
+#    define waitOutputPinValid() wait_cpuclock(WAITOUTPUTPINVALID_DELAY)
 #endif
 
 // Atomic macro to help make GPIO and other controls atomic.
diff --git a/tmk_core/common/wait.h b/tmk_core/common/wait.h
index 7d0c0e3b70a4..447bb9275883 100644
--- a/tmk_core/common/wait.h
+++ b/tmk_core/common/wait.h
@@ -9,7 +9,7 @@ extern "C" {
 
 #if defined(__ARMEL__) || defined(__ARMEB__)
 #    ifndef __OPTIMIZE__
-#        warning "Compiler optimizations disabled; wait_cpuclock() won't work as designed"
+#        pragma message "Compiler optimizations disabled; wait_cpuclock() won't work as designed"
 #    endif
 
 #    define CLOCK_DELAY_NOP8 "nop\n\t nop\n\t nop\n\t nop\n\t   nop\n\t nop\n\t nop\n\t nop\n\t"

From b16caee3c7ee1e14f3352107e813b1f3c60def38 Mon Sep 17 00:00:00 2001
From: mtei <2170248+mtei@users.noreply.github.com>
Date: Sat, 14 Nov 2020 20:01:50 +0900
Subject: [PATCH 13/17] waitOutputPinValid() rename to waitInputPinDelay() in
 quantum/quantum.h.

---
 quantum/quantum.h | 28 ++++++++++++++++++++++------
 1 file changed, 22 insertions(+), 6 deletions(-)

diff --git a/quantum/quantum.h b/quantum/quantum.h
index 7f13cc7cfa9e..ea1432ba2f01 100644
--- a/quantum/quantum.h
+++ b/quantum/quantum.h
@@ -208,7 +208,11 @@ typedef uint8_t pin_t;
 
 #    define togglePin(pin) (PORTx_ADDRESS(pin) ^= _BV((pin)&0xF))
 
-#    define waitOutputPinValid() wait_cpuclock(1)
+/*   The AVR series GPIOs have a one clock read delay for changes in the digital input signal. */
+#    if !defined(GPIO_INPUT_PIN_DELAY)
+#        define GPIO_INPUT_PIN_DELAY 1
+#    endif
+#    define waitInputPinDelay() wait_cpuclock(GPIO_INPUT_PIN_DELAY)
 
 #elif defined(PROTOCOL_CHIBIOS)
 typedef ioline_t pin_t;
@@ -229,12 +233,24 @@ typedef ioline_t pin_t;
 #endif
 
 #if defined(__ARMEL__) || defined(__ARMEB__)
-#    if  defined(STM32_SYSCLK)
-#        define WAITOUTPUTPINVALID_DELAY (STM32_SYSCLK/1000000L / 4)
-#    elif defined(KINETIS_SYSCLK_FREQUENCY)
-#        define WAITOUTPUTPINVALID_DELAY (KINETIS_SYSCLK_FREQUENCY/1000000L / 4)
+/* For GPIOs on ARM-based MCUs, the input pins are sampled by the clock of the bus
+ * to which the GPIO is connected.
+ * The connected buses differ depending on the various series of MCUs.
+ * And since the instruction execution clock of the CPU and the bus clock of GPIO are different,
+ * there is a delay of several clocks to read the change of the input signal.
+ *
+ * Define this delay with the GPIO_INPUT_PIN_DELAY macro.
+ * If the GPIO_INPUT_PIN_DELAY macro is not defined, the following default values will be used.
+ * (A fairly large value of 0.25 microseconds is set.)
+ */
+#    if !defined(GPIO_INPUT_PIN_DELAY)
+#        if defined(STM32_SYSCLK)
+#            define GPIO_INPUT_PIN_DELAY (STM32_SYSCLK/1000000L / 4)
+#        elif defined(KINETIS_SYSCLK_FREQUENCY)
+#            define GPIO_INPUT_PIN_DELAY (KINETIS_SYSCLK_FREQUENCY/1000000L / 4)
+#        endif
 #    endif
-#    define waitOutputPinValid() wait_cpuclock(WAITOUTPUTPINVALID_DELAY)
+#    define waitInputPinDelay() wait_cpuclock(GPIO_INPUT_PIN_DELAY)
 #endif
 
 // Atomic macro to help make GPIO and other controls atomic.

From c4e40c7cd6046acd51b67fea145eee5e96977032 Mon Sep 17 00:00:00 2001
From: mtei <2170248+mtei@users.noreply.github.com>
Date: Sat, 14 Nov 2020 20:08:55 +0900
Subject: [PATCH 14/17] waitOutputPinValid() rename to waitInputPinDelay() in
 quantum/matrix_common.c

---
 quantum/matrix_common.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/quantum/matrix_common.c b/quantum/matrix_common.c
index d5a857084464..01d2b38e5e51 100644
--- a/quantum/matrix_common.c
+++ b/quantum/matrix_common.c
@@ -87,7 +87,7 @@ uint8_t matrix_key_count(void) {
 /*　`matrix_io_delay ()` exists for backwards compatibility. From now on, use matrix_output_unselect_delay().　*/
 __attribute__((weak)) void matrix_io_delay(void) { wait_us(MATRIX_IO_DELAY); }
 
-__attribute__((weak)) void matrix_output_select_delay(void) { waitOutputPinValid(); }
+__attribute__((weak)) void matrix_output_select_delay(void) { waitInputPinDelay(); }
 __attribute__((weak)) void matrix_output_unselect_delay(void) { matrix_io_delay(); }
 
 // CUSTOM MATRIX 'LITE'

From 7fb62934a9bc59f17fc67e94f437724f831d3f44 Mon Sep 17 00:00:00 2001
From: mtei <2170248+mtei@users.noreply.github.com>
Date: Mon, 16 Nov 2020 17:15:10 +0900
Subject: [PATCH 15/17] update tmk_core/common/wait.h

---
 tmk_core/common/wait.h | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/tmk_core/common/wait.h b/tmk_core/common/wait.h
index 447bb9275883..e224ff38ca16 100644
--- a/tmk_core/common/wait.h
+++ b/tmk_core/common/wait.h
@@ -12,12 +12,12 @@ extern "C" {
 #        pragma message "Compiler optimizations disabled; wait_cpuclock() won't work as designed"
 #    endif
 
-#    define CLOCK_DELAY_NOP8 "nop\n\t nop\n\t nop\n\t nop\n\t   nop\n\t nop\n\t nop\n\t nop\n\t"
+#    define wait_cpuclock(x) wait_cpuclock_allnop(x)
 
-#define WAIT_CPUCLOCK_DEFINED
+#    define CLOCK_DELAY_NOP8 "nop\n\t nop\n\t nop\n\t nop\n\t   nop\n\t nop\n\t nop\n\t nop\n\t"
 
 __attribute__((always_inline))
-static inline void wait_cpuclock(unsigned int n) { /* n: 1..135 */
+static inline void wait_cpuclock_allnop(unsigned int n) { /* n: 1..135 */
     /* The argument n must be a constant expression.
      * That way, compiler optimization will remove unnecessary code. */
     if (n < 1) { return; }

From 37744309075da3e0bb513058199817ecc26b9dea Mon Sep 17 00:00:00 2001
From: mtei <2170248+mtei@users.noreply.github.com>
Date: Mon, 16 Nov 2020 17:17:09 +0900
Subject: [PATCH 16/17] update comment in quantum/matrix.c,
 quantum/split_common/matrix.c

---
 quantum/matrix.c              | 2 +-
 quantum/split_common/matrix.c | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/quantum/matrix.c b/quantum/matrix.c
index 8778fb036477..2eec1259831f 100644
--- a/quantum/matrix.c
+++ b/quantum/matrix.c
@@ -189,7 +189,7 @@ static bool read_rows_on_col(matrix_row_t current_matrix[], uint8_t current_col)
     // Unselect col
     unselect_col(current_col);
     if (current_col + 1 < MATRIX_COLS) {
-        matrix_output_unselect_delay();  // wait for row signal to go HIGH
+        matrix_output_unselect_delay();  // wait for col signal to go HIGH
     }
 
     return matrix_changed;
diff --git a/quantum/split_common/matrix.c b/quantum/split_common/matrix.c
index 5d2c4caba11e..24fb2ae53b03 100644
--- a/quantum/split_common/matrix.c
+++ b/quantum/split_common/matrix.c
@@ -202,7 +202,7 @@ static bool read_rows_on_col(matrix_row_t current_matrix[], uint8_t current_col)
     // Unselect col
     unselect_col(current_col);
     if (current_col + 1 < MATRIX_COLS) {
-        matrix_output_unselect_delay();  // wait for row signal to go HIGH
+        matrix_output_unselect_delay();  // wait for col signal to go HIGH
     }
 
     return matrix_changed;

From 9ce7c0a35e487d7450ab2c6897bd2c91e0c75362 Mon Sep 17 00:00:00 2001
From: mtei <2170248+mtei@users.noreply.github.com>
Date: Mon, 23 Nov 2020 15:12:20 +0900
Subject: [PATCH 17/17] update quantum/quantum.h: Make more margin in the
 GPIO_INPUT_PIN_DELAY default value.

---
 quantum/quantum.h | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/quantum/quantum.h b/quantum/quantum.h
index ea1432ba2f01..60523feebdae 100644
--- a/quantum/quantum.h
+++ b/quantum/quantum.h
@@ -208,9 +208,10 @@ typedef uint8_t pin_t;
 
 #    define togglePin(pin) (PORTx_ADDRESS(pin) ^= _BV((pin)&0xF))
 
-/*   The AVR series GPIOs have a one clock read delay for changes in the digital input signal. */
+/*   The AVR series GPIOs have a one clock read delay for changes in the digital input signal.
+ *   But here's more margin to make it two clocks. */
 #    if !defined(GPIO_INPUT_PIN_DELAY)
-#        define GPIO_INPUT_PIN_DELAY 1
+#        define GPIO_INPUT_PIN_DELAY 2
 #    endif
 #    define waitInputPinDelay() wait_cpuclock(GPIO_INPUT_PIN_DELAY)