Micro-optimizations

1 files changed, 23 insertions, 30 deletions

diff --git a/dla.s b/dla.s
index 691387f..1886944 100644
--- a/dla.s
+++ b/dla.s
@@ -502,28 +502,27 @@ spawn:
 ;;; This is the innermost loop, so it should be as optimized as
 ;;; possible (we're not there yet).
 drunkwalk:
- lda RANDOM ; pick a random direction, up/down/left/right
- and #$C0   ; use top 2 bits (hopefully more random than bottom 2).
- cmp #$C0
- beq up
- cmp #$80
- beq down
- cmp #$40
- beq left
- ; $00: right
- inc part_x
- bne checkbounds
-up:
- dec part_y
- bne checkbounds
+ ; using bit/bmi/bvc saves 6.25 cycles on average, compared to
+ ; immediate cmp and bne.
+ ; 4 code paths: up=14, down=15, left=15, right=13, avg=14.25
+ bit RANDOM ;4 ; use top 2 bits (probably more random, definitely faster)
+ bmi lr     ;2/3
+ bvc down   ;2/3
+ dec part_y ;3 ; N=1 V=1 up
+ bne checkbounds ;3
 down:
- inc part_y
- bne checkbounds
-left:
- dec part_x
+ inc part_y ;3 ; N=1 V=0 down
+ bne checkbounds ;3
+lr:
+ bvc right ;2/3
+ dec part_x ;3 ; N=0 V=1 left
+ bne checkbounds ;3
+right:
+ inc part_x ;3
 
 checkbounds:
  lda part_x
+ sta cursor_x
  cmp min_x
  beq oob
  cmp max_x
@@ -534,10 +533,8 @@ checkbounds:
  cmp max_y
  beq oob
 
- lda part_x
- sta cursor_x
- lda part_y
  sta cursor_y
+
  ldx #0
  lda CONSOL
  cmp #6
@@ -556,23 +553,19 @@ dontplot:
  and (pixptr),y
  bne stick
  ; (1,0)
- inc cursor_x
- inc cursor_x
- jsr plotsetup
+ inx
+ stx cursor_x
+ inx
+ ldy xoffsets,x
+ lda xmasks,x
  and (pixptr),y
  bne stick
  ; (0,-1)
- dec cursor_x
  dec cursor_y
  jsr plotsetup
  and (pixptr),y
  bne stick
  ; (0,1)
- ; used to:
- ;inc cursor_y
- ;inc cursor_y
- ;jsr locate
- ; this avoids recalculating the pointer:
  tya
  ora #$40 ; add 64
  tay