Tummai Games

Took some time away from work and hobbies to spend some time with the family. Before I left I was working on adding some enemies into the game. I made three different monsters and added enemy-loading logic to the room-loading routines. The enemies display onscreen fine and die when you shoot their HP to 0. Tomorrow I’m going to work on some movement patterns.

I felt I should do some kind of update though, so here are the terrifying monsters I whipped up in GIMP. Turn on the lights before looking at them or you might die in your sleep:

1. Purple Tongue Head
   
   

   When you see it coming, the shit runs down your leg.

2. Eyeball
   
   

   It's looking for something to KILL!

3. Shuriken
   
   

   It's pointy.

I haven’t posted in a few days and that’s because I’ve been spending my time doing research. I’ve been peeking at other games and seeing how they handle sprite objects. By sprite object, I mean entities in a game that are represented on screen with sprites: enemies, projectiles, powerups, etc. When I was trying to code bullets, I didn’t really know how to go about it. Should I keep them separate from other sobjects (sprite objects) or are there enough similarities to lump them together?

If you think about it, a bullet isn’t much different from an enemy. It has an x and a y. It has a speed and a movement pattern. It has a palette. It has a hit box. The main difference is what you check for collision against. Enemies collide with the hero. Hero bullets don’t. They collide with enemies instead.

So some logic is different, but the data (or rather type of data stored) is the same! As far as turning the sobject data into actual sprites, bullets can be treated just the same as enemies (the hero too). Why do I care? If I am able to lump them all together, I can make all the sprites in a single loop. Here’s an example of what I mean:

Sobjects to Sprites

Let’s say I have a hero. That’s one sobject. He can shoot bullets. Let’s say I want to allow 3 bullets maximum on the screen at a time. That’s three more potential sobjects. Then enemies. Just pulling a number out of the air, lets say I want there to be 10 enemies on the screen at a time maximum. That’s ten more potential sobjects.

1 hero + 3 bullets + 10 enemies = 14.

So I have 14 sobjects that could be represented on the screen at any one time. If I treat them all the same, I can hold all of the data in arrays, like this:

sobj_id: .res 14
sobj_hp: .res 14
sobj_x: .res 14
sobj_y: .res 14
sobj_hitbox_top: .res 14
sobj_hitbox_bottom: .res 14
...etc

Then whenever I need to turn them into sprites, I can just loop through them all using an index register:

sobjects_to_sprites:
    ldx #$00
@loop:
    lda sobj_id, x
    ;do stuff
    ;calculate y, CHR#, attrib and x
    ;for each sprite of the sprite object
    ;and store in DMA sprite RAM

    inx
    cpx #.sizeof(sobj_id) ;loop through all 14
    bne @loop
    rts

And that will cover all my sobjects as far as turning their data into sprites. I don’t need to write more than one routine!

To differentiate between sobject types (bullet vs. enemy vs. hero), I can just have their index range set. For example, hero is always index 0. Bullets are always 1-3. And enemies are always 4-13. This way I always know what to test for collisions against: enemies will always test against index 0 (hero) and deal damage on a collision. Enemies will also test against index 1-3 (bullets) and receive damage on a collision.

With defined index ranges, limit testing is easy too: If the player presses the fire button, check sobjects 1-3. If they are full already, don’t create a new bullet. If there is an open space there, create a new bullet.

Conclusion

I think I finally figured out how I’m going to arrange sprite objects. Today I rewired all of my previous hero-moving code to fit into this new data model and I think I have it working. Tomorrow I’m going to try to add bullets for real!

I originally started working on this game as an entry for a coding competition. The genre was puzzle games and I thought a switch-throwing maze game would be fun to write. I missed the deadline for the contest long ago but I kept working on the game anyway. Now I’m thinking about taking the game in a new direction.

The decision to make a maze-only game was made because there was a deadline, and because the genre for the contest was “puzzle”. But now that I have as much time and freedom as I want, I think I’m going to expand the game beyond just finding your way through a maze. Specifically, here are some things I want to add that weren’t in the original plan:

1. Enemies and bosses
2. Weapons
3. A way to die (ie, health bar)
4. A way to improve (more weapons. powerups… level ups?)

In addition to these changes, I’m also going to change the setting and storyline completely. No more treasure hunting archaeologist. I haven’t worked out all the details but I think it’s going to be somewhat sci-fi, with the hero as a cyborg (ie, guns not swords). I think using the Guardian Legend girl as my test sprite has influenced me in some way. :).

Anyway, this comes at just the right time. I just got switches and doors working and the next thing I was going to write was a way to break the breakable blocks. In the original maze-game concept there was going to be an item (like a hammer) that the archaeologist hero would find to allow him to break blocks. Now I think it will be a powerup that makes your gun blast stronger.

Next objective: make the hero shoot a projectile when pressing A.

In the last post, I got the hero sprite moving. But there’s a problem: she can walk through walls. I need to add collision detection with background tiles. Collision detection with the background can be summed up like this:

if the player tries to move:
    check the space they want to move to
    if that space is solid/blocked:
        don't move
    else:
        move

The looks pretty easy, but it can actually be quite complicated. Here are some issues:

1. The hero sprite has x and y coordinates, which are on the pixel level. My room tile data is stored in ram on the metatile level. We need to make a conversion there.
2. There isn’t any one point on the hero that is a catch-all for movement checks. If the player tries to move left, we will want to check for a collision with the left side of the hero. If the player tries to move right, we want to check for a collision with the right side of the hero. So we will need to make a hit box.
3. Each side of the hit box will have a length > 1, so we may have to check for a collision with more than one background tile. For example, say the player wants to move right. We will check using the right side of the hit box. But it’s possible that the top half of the hit box will be adjacent to one tile, while the bottom half is adjacent to another.
   

   If we only checked the top-right corner of this hit box for collision, it would show the way as clear. But it really isn't, because the bottom half of the hit box collides with a solid tile.

Step 1

First thing I want to do is make a hit box. I’m going to put it around the hero sprite’s feet. I define where the edges are relative to the hero’s X/Y coords, then make a subroutine to calculate the box every frame:

;collision box for the hero sprite (box around the feet)
HERO_MOVEMENT_BOX_TOP = 26
HERO_MOVEMENT_BOX_BOTTOM = 31
HERO_MOVEMENT_BOX_LEFT = 5
HERO_MOVEMENT_BOX_RIGHT = 11

hero_box: .res 4 ;top bottom left right

set_hero_box:
    lda hero_x
    clc
    adc #HERO_MOVEMENT_BOX_LEFT
    sta hero_box+2
    lda hero_x
    clc
    adc #HERO_MOVEMENT_BOX_RIGHT
    sta hero_box+3
    lda hero_y
    clc
    adc #HERO_MOVEMENT_BOX_TOP
    sta hero_box
    lda hero_y
    clc
    adc #HERO_MOVEMENT_BOX_BOTTOM
    sta hero_box+1
    rts

Movement within a metatile

Since every metatile is 16 pixels wide and 16 pixels tall, I can take a shortcut if I determine that the player is not at the edge of a tile. Check out these two cases:

We don't need to check for a collision in the left case. The player is already on a walkable tile, and will remain on that tile if they move a pixel to the right.

Let’s assume the player is trying to move right. The player is on a walkable tile in both cases. In the first case, they are in the middle of the tile. In the second case they are at the edge. We only need to check for a collision in the second case. Checking the first case would be a waste of time, because we’d be checking the tile the player is already on for walkability, but we know it must be walkable since the player is already on it. So right away we can skip over a collision check if we determine the player to be in the middle of the tile (ie, not on the edge).

This is very easy to pull off with 16×16 metatiles. In all cases, the left edge of a metatile will have an x-coord of $x0, and the right edge will be $xF. The top edge of a metatile will have a y-coord of $x0 and the bottom will be $xF. So in our collision detection routines, we can do something like this:

can_move_right:
    ldx hero_box+3 ;right edge of movement box
    inx ;we want to peek at the next pixel over
    txa

    and #$0F ;isolate right nibble,
         ; tells us where we are WITHIN a tile

    bne @move_ok ;not on the edge.  movement within current
                 ;square OK (in other words, if the right
                 ;nibble is 0, we are on the left edge of the
                 ;NEXT tile, and therefore need to check
                 ;for collision)

    ;... check for collisions

@move_ok:
    sec  ;return a 1 if movement is OK
    rts

I use a similar check at the beginning of can_move_left, can_move_up and can_move_down.

How many checks?

Do I need to check one bg tile for collision or two? It depends on the hero sprite’s position. Imagine the player is moving right again. If the top and bottom edges of the hit box are both within the boundaries of a single tile, we only need to check one bg tile for collision. But check this picture (should look familiar):

moving right, the top and bottom edges of the hit box touch two different bg tiles. We need to make two checks.

In this case, the top and bottom edges of our hit box line up with two different bg tiles. So we need to check both. If either of them are unwalkable, we don’t allow movement.

So, two checks or one check? There are a couple of ways to determine if the top and bottom edges of our hit box touch different tiles. I do it this way:

    ;find how many tiles to check (1 or 2)
    lda hero_box ;top edge of movement hit box

    ora #$F0 ;we are going to add the box height,
             ;so set us up to check for a FF->00 transition

    adc #HERO_MOVEMENT_BOX_HEIGHT ;negative means our box is
                                  ;h-aligned with one tile.
                                  ;positive = 2 tiles
    bmi @not_two

    ;... check a tile

@not_two:
    ;... check a tile

I do similar tests for the other directions, but for can_move_up and can_move_down I will check the left and right edges instead of the top and bottom edges.

Hero pixel coords to a room coords

I need to turn the hero’s pixel coordinates into room coordinates so that I can find which bg tiles to check. Here’s my subroutine to do that:

;------------------------------
; set_hero_map_coords finds the x and y room coords for the 
; topleft pixel of the player's movement hit box
set_hero_map_coords:
    lda hero_box+2 ;left
    ldx #$00
    sec
:
    sbc #$10
    bcc :+
    inx
    jmp :-
:
    stx hero_map_x

    lda hero_box ;top
    ldx #$00
    sec
:
    sbc #$10
    bcc :+
    inx
    jmp :-
:
    dex
    dex ;correct y for the status bar
    stx hero_map_y
    rts

Once I have the hero’s room coordinates, I can add to or subtract from them to find the coordinates for the adjacent tiles I want to check for walkability.

Walkability

How do I check for walkability? With a lookup table:

;tile ids
.enum
    floor
    wall
    block
    water
    block_breakable
    stairs_up
    stairs_down
    pitfall
.endenum

.enum ;walkability
    unwalkable
    walkable
.endenum

tile_walkability:
    .byte walkable, unwalkable, unwalkable, unwalkable
    .byte unwalkable, walkable, walkable, walkable

I read a tile id from the room map in RAM and use it to index into the tile_walkability table.

Here’s a simplified can_move_right. (I took out checks for room boundaries to make it more readable):

can_move_right:
    ldx hero_box+3 ;right edge of movement box
    inx
    txa

    and #$0F ;isolate right nibble, where are we WITHIN a tile?

    bne @move_ok ;not on edge. move within current square OK

    ;how many tiles to check (1 or 2)
    lda hero_box ;top edge of movement box
    ora #$F0
    adc #HERO_MOVEMENT_BOX_HEIGHT
    bmi @not_two

    ldy hero_map_y
    iny
    ldx hero_map_x
    inx
    jsr get_room_offset ;takes x/y room coords and returns an
                        ;array index in y

    lda room, y
    tay
    lda tile_walkability, y
    beq @no_move

@not_two:
    ldy hero_map_y
    ldx hero_map_x
    inx
    jsr get_room_offset

    lda room, y
    tay
    lda tile_walkability, y
    beq @no_move
@move_ok:
    sec return 1 in the carry if we can move
    rts
@no_move:
    clc return 0 in the carry if we can't move
    rts

I have seperate routines for the other 3 directions. They are all very similar. The last step is to update my move_hero subroutine to call these collision detection routines before moving, and then skip movement on carry clear.

Conclusion

Now I have a sprite that changes direction, moves, animates and bumps into walls. Here’s the latest video demo:

Ok, I got the hero to change directions based on user input. Now it’s time to move her. The first step for this is to have a hero_moving flag that the input handler will set if there is d-pad input:

hero_moving: .res 1

handle_joypad:
    lda joypad1
    and #$F0
    beq @end
    lsr
    lsr
    lsr
    lsr
    tay
    lda direction_change_table, y
    sta hero_direction
    lda #$01
    sta hero_moving
    rts
@end:
    lda #$00
    sta hero_moving
    rts

With this updated input handler, the hero_moving flag will be 1 when the d-pad is pressed, and 0 when it is not. And as before, our hero’s direction will be stored in hero_direction.

Next, I need to check the hero moving flag in my update_hero_sprite routine, which is called every frame:

update_hero_sprite:
    lda hero_moving
    beq @hero_moving_done

    jsr move_hero
@hero_moving_done:
    ;...
    ;write sprite data to RAM as before

Now that I have the foundation laid, I need to write the move_hero routine.

move_hero

There are 8 possible directions the hero could be facing. The current direction is stored in a variable called hero_direction. The possible values for this variable are 0-7, as assigned by these constants:

;direction indexes for tables
SPRITE_UP =         $00
SPRITE_DOWN =       $01
SPRITE_LEFT =       $02
SPRITE_RIGHT =      $03
SPRITE_UP_LEFT =    $04
SPRITE_UP_RIGHT =   $05
SPRITE_DOWN_LEFT =  $06
SPRITE_DOWN_RIGHT = $07

I need to alter the hero’s coordinates based on the direction the player is moving. If they are moving left, I need to subtract from hero_x. If they are moving right, I need to add to hero_x. If they are moving up, I need to subtract from hero_y. If they are moving down, I need to add to hero_y. If they are moving diagonally, I need to update both hero_x and hero_y. There are many possibilities, and rather than have a million branch instructions, I’ll use table lookups again. I will have one table for horizontal movement and one table for vertical movement. I will index into these tables using hero_direction. The tables look like this:

;up, down, left, right, up_left, up_right, down_left, down_rt
hero_movement_x:
    .byte $00, $00, $FF, $01, $FF, $01, $FF, $01

hero_movement_y:
    .byte $FF, $01, $00, $00, $FF, $FF, $01, $01

My move_hero routine will read from these tables and add to the hero coordinates. Note that adding $FF to a number (on an 8-bit system) is the same as subtracting 1, since it will wrap around from $FF to $00. Here is move_hero:

move_hero:
    ldy hero_direction
    lda hero_movement_x, y
    clc
    adc hero_x
    sta hero_x

    ldy hero_direction
    lda hero_movement_y, y
    clc
    adc hero_y
    sta hero_y

    rts

And now she’s moving.

miau pointed out in a comment that diagonal movement should really alter x and y by SQRT(2) rather than 1 (see Pythagorean Theorum). This is something I didn’t consider before, so my hero covers ground a little more quickly when moving diagonally. I’m not sure yet if this is undesirable behavior or not. If I keep it a simple maze-solving game, it probably won’t be an issue. If I get more ambitious and change it to a more action-oriented game with enemies (something I’ve been seriously thinking about), it may be an issue. We’ll see :).

Animation

I want to talk very briefly about animating the sprite. To do animation, I need a few things:

1. Graphics for the various frames of animation
2. A counter to tell me when to change animation frames
3. A variable telling me the current animation frame

I already have the graphics for my test sprite. The other two I need to make myself.

frame_counter: .res 1
hero_anim_frame: .res 1

The animation is a walking animation, so I don’t want the hero to animate unless she is moving. It makes sense then to do the animation frame-changing logic in the move_hero routine. The Guardian Legend hero sprite has a 4-frame walking animation, so here’s how I do it:

move_hero:
    ldy hero_direction
    lda hero_movement_x, y
    clc
    adc hero_x
    sta hero_x

    ldy hero_direction
    lda hero_movement_y, y
    clc
    adc hero_y
    sta hero_y

    inc frame_counter
    lda frame_counter
    cmp #$0B ;change anim frame every 11 "moving frames"
             ;found this number by trial and error
    bcc @end

    lda #$00
    sta frame_counter ;reset counter
    inc hero_anim_frame ;go to next anim frame
    lda hero_anim_frame ;make sure we stay between 0 and 3
    and #$03
    sta hero_anim_frame
@end:
    rts

Then I will modify my CHR tile lookup tables to handle graphics for all four frames of animation for all 8 directions. Then I will update my update_hero_sprite routine to read from these tables based on the values of hero_anim_frame and hero_direction. I’m not going to post the code, because it’s a little long and there are a lot of tables. If anybody wants me to elaborate more, let me know in the comments :)

BTW, before I forget! When adding new variables, it’s a good idea to initialize them. Here is my updated initialize_hero_sprite routine:

initialize_hero_sprite:
    lda #$50
    sta hero_x
    sta hero_y

    lda #$00
    sta hero_moving
    sta hero_anim_frame
    sta frame_counter

    lda #SPRITE_DOWN
    sta hero_direction

    jsr update_hero_sprite

    rts

Conclusion

Whew. I covered a lot of ground today. I probably should have made this two posts instead of one, but I want to hurry up and get to collision detection so I can put up the next demo video. :)

You may have noticed by now that I use hero_direction a lot to index into lookup tables. This is because everything changes based on the hero’s direction, and lookup tables are a great alternative to long sections of compare/branch code. Often when I find myself writing a lot of branching code I’ll see if I can come up with some scheme to turn the test value into a table index.

See you next time!