Some missed stuff from that aborted fps-count patch

Author: ben-xo

fps_count.h

@@ -17,6 +17,20 @@
 // this is defined in fps.cpp
 extern int8_t volatile fps_interrupt_count;
 
+/*
+ * Every time fps_count() is called, it decrements a counter of "interrupts per frame".
+ *
+ * We hitch a ride on the LED PWM interrupt (see ledpwm.cpp), as the timing there
+ * is essentially jitter free.
+ *
+ * For example, if our desired frame rate is 150FPS, and our interrupt triggers 10k times
+ * per second (i.e. a typical LED PWM value), then we count down from 200 to 0 and set a
+ * flag in GPIOR0 when it underflows. (See PWM_OVERFLOW_VALUE in ledpwm.cpp)
+ *
+ * GPIOR0 is chosen because you can use `sbi` to set the bit very quickly.
+ *
+ * This whole function only takes 7 or 8 clock cycles.
+ */
 static void inline __attribute__((always_inline)) fps_count()
 {
   // clobbers r24 and SREG so they must be saved before use.

ledpwm.cpp

@@ -117,14 +117,14 @@ ISR(TIMER2_COMPB_vect, ISR_NAKED) {
   register bool is_beat_2 asm ("r24") = F.is_beat_2; // 2cy
   if(is_beat_2) tempo_pin.high(); // this compiles to a `sbrc` which doesn't affect the SREG!
 
-  /* Based on LSB of portb_mask, swap the nibbles of portb val before di
-   * The idea is that portb_val is actually a double buffer, and portb_m
-   * a blend percentage. Once every sample interrupt, it is rotated by 1
-   * So a mask = 0x00 will always show the one half of portb_val, and ma
-   * with mask = 0x55 showing a 50/50 mix. Thus, you can achieve fades a
+  /* Based on LSB of portb_mask, swap the nibbles of portb val before displaying.
+   * The idea is that portb_val is actually a double buffer, and portb_mask is effectively
+   * a blend percentage. Once every sample interrupt, it is rotated by 1 bit.
+   * So a mask = 0x00 will always show the one half of portb_val, and mask = 0xFF will show the other half,
+   * with mask = 0x55 showing a 50/50 mix. Thus, you can achieve fades and pulses on the seven seg
    * by periodically updating the val and the mask.
    *
-   * Note that because just saving and restoring SREG takes 6 cycles, we
+   * Note that because just saving and restoring SREG takes 6 cycles, we're avoiding anything that modifies
    * SREG altogether in order to keep this to 8 cycles total.
    *
    * total: 8 cycles