back porch skipping, wasting clock cycles, basically

const.h

@@ -1,4 +1,4 @@
 #pragma once
 
 #define VERTICAL_OFFSET 30
-#define HORIZONTAL_OFFSET_CYCLE 50
+#define HORIZONTAL_OFFSET_CYCLE 56

main.S

@@ -36,24 +36,27 @@ main:
     rjmp 1b
 
 .global int_horizontal_sync
-int_horizontal_sync:
-    push r31
-    in r31, IO(SREG)    ; status register
-    push r31
-    push r30
+int_horizontal_sync:    ; +3
+    push r31            ; +5
+    in r31, IO(SREG)    ; +6, status register
+    push r31            ; +8
+    push r30            ; +10
 
     ; if (line >= VERTICAL_OFFSET), then enter
-    lds r30, line
-    lds r31, line + 1
-    adiw z, 1
-    sts line, r30
-    sts line + 1, r31
-    cpi r31, 0
-    brne enter
-    cpi r30, VERTICAL_OFFSET
-    brlo int_horizontal_sync_end
+    lds r30, line       ; +12
+    lds r31, line + 1   ; +14
+    adiw z, 1           ; +16
+    sts line, r30       ; +18
+    sts line + 1, r31   ; +20
+    cpi r31, 0          ; +21
+    brne enter                      ; +22, +23
+    cpi r30, VERTICAL_OFFSET        ; +23
+    brlo int_horizontal_sync_end    ; +24
 
 enter:
+    ; here, +23 or +24 cycles have passed since horizontal sync
+    ; so, there are still ~168 cycles before first useful data
+    ; or 56 3-cycle instructions (HORIZONTAL_OFFSET_CYCLE)
     ldi r31, HORIZONTAL_OFFSET_CYCLE    ; skip back porch
 1:
     dec r31