Start work on sprite attribute caching

demos/pacman/App.s

@@ -1,10 +1,12 @@
-; IIgs Game Engine
+; IIgs Shooter
 
             TYP   $B3         ; S16 file
-            DSK   GTEPacMan
+            DSK   GTEShooter
             XPL
 
 ; Segment #1 -- Main execution block
 
             ASM   Main.s
 
+
+

demos/smb/App.s

@@ -0,0 +1,61 @@
+; IIgs Game Engine
+
+            TYP   $B3         ; S16 file
+            DSK   GTETestApp
+            XPL
+
+; Segment #1 -- Main execution block
+
+            ASM   Main.s
+
+; Segment #2 -- ROM
+
+            ASM   rom.s
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+

demos/smb/Main.s

@@ -0,0 +1,4 @@
+            REL
+            DSK   MAINSEG
+
+            rtl

demos/smb/wrapper.s

@@ -0,0 +1,12 @@
+; Wrapper around the SMB ROM code.  This takes care of setting any memory locations
+; in the SMB ROM address space that interact with the system (like controller inputs).
+;
+; We also shim many of the LDA/STA instruction that modify the NES I/O to be
+; JSRs to small subroutines that enqueue any changes that are handled once 
+; control returns.  The queues are important, because we try to run the game
+; logic at 60 fps, but the screen will update significantly slower than that.
+;
+; By queuing the changes, we can "catch up" to the game logic and prioritize
+; audio output at 60 fps since audio stutter is much more disruptive that slow
+; FPS.
+

demos/sprites/gen/App.TileMapBG0.s

@@ -208,7 +208,7 @@ App_TileMapBG0
         dw $1021,$1021,$1031,$1032,$1021,$1021,$100f,$0010,$0011,$1012,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$100f,$0010,$0011,$1012,$1021,$1021,$0007,$0008,$0007,$0008,$0007,$0008,$0007,$0008,$0007,$0008,$0007,$0008,$0007,$0008,$0007,$0008,$1021,$1021,$1021,$1021
         dw $1021,$1021,$1030,$0015,$0015,$0015,$0015,$1033,$1021,$1021,$1021,$1021,$1009,$100a,$1021,$1021,$1021,$1021,$1021,$1021,$001a,$001a,$001a,$001a,$0040,$0040,$001a,$001a,$001a,$001a,$1021,$1021
 
-        dw $1021,$1030,$0015,$0036,$0015,$0015,$0036,$0015,$1033,$1021,$1800,$1804,$0800,$0804,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1022,$1023,$1022,$1023,$1022,$1023,$1021,$1021,$1021,$1030,$0015,$0036,$1230,$1021,$1021,$1057,$1056,$1057,$1056,$1057,$1056,$1057,$1056,$1021,$1022,$1023,$1021,$1021,$1057,$1056,$1057,$1056,$1013,$0014,$0015,$1016,$1021,$1021,$1021,$1021
+        dw $1021,$1030,$0015,$0036,$0015,$0015,$0036,$0015,$1033,$1021,$1800,$1804,$0800,$0804,$5800,$5804,$4800,$4804,$1021,$1021,$1021,$1021,$1021,$1021,$1022,$1023,$1022,$1023,$1022,$1023,$1021,$1021,$1021,$1030,$0015,$0036,$1230,$1021,$1021,$5057,$5056,$5057,$5056,$1057,$1056,$1057,$1056,$1021,$1022,$1023,$1021,$1021,$1057,$1056,$1057,$1056,$1013,$0014,$0015,$1016,$1021,$1021,$1021,$1021
         dw $1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1013,$0014,$0015,$1016,$1021,$1021,$1021,$1021,$1022,$1023,$1022,$1023,$1021,$1021,$1021,$1021,$1013,$0014,$0015,$1016,$1021,$1030,$0015,$0036,$0015,$0015,$0036,$0015,$1033,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1013,$0014,$0015,$1016,$1021,$1021,$1022,$1023,$1022,$1023,$1022,$1023,$1021,$1021
         dw $1021,$1030,$0015,$0036,$1033,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1022,$1023,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1022,$1023,$1022,$1023,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021
         dw $1021,$1030,$0015,$0036,$0015,$0015,$0036,$0015,$1033,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1022,$1023,$1022,$1023,$1022,$1023,$1021,$1021,$1021,$1030,$0015,$0036,$1033,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1022,$1023,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021
@@ -216,7 +216,7 @@ App_TileMapBG0
         dw $1021,$1030,$0015,$0036,$1033,$1021,$1013,$0014,$0015,$1016,$1021,$1021,$1021,$1021,$0000,$1021,$1022,$1023,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1013,$0014,$0015,$1016,$0005,$0006,$0005,$0006,$0005,$0006,$0005,$0006,$0005,$0006,$0005,$0006,$0005,$0006,$0005,$0006,$0005,$0006,$1021,$1021,$1021,$1021
         dw $1021,$1030,$0015,$0036,$0015,$0015,$0036,$0015,$1033,$1021,$1021,$1021,$0005,$0006,$1021,$1021,$1021,$1021,$1021,$1021,$001a,$001a,$001a,$001a,$0040,$0040,$001a,$001a,$001a,$001a,$1021,$1021
 
-        dw $1030,$0015,$0015,$0015,$0015,$0015,$0015,$0015,$0015,$1033,$1808,$180C,$0808,$080C,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1024,$0025,$0025,$0025,$0025,$0025,$0025,$1026,$1021,$1030,$0015,$0015,$0015,$0015,$1230,$1021,$1077,$1076,$1077,$1076,$1077,$1076,$1077,$1076,$1024,$0025,$0025,$1026,$1021,$1077,$1076,$1077,$1076,$1013,$0014,$0015,$1016,$1021,$1021,$1021,$1021
+        dw $1030,$0015,$0015,$0015,$0015,$0015,$0015,$0015,$0015,$1033,$1808,$180C,$0808,$080C,$5808,$580C,$4808,$480C,$1021,$1021,$1021,$1021,$1021,$1024,$0025,$0025,$0025,$0025,$0025,$0025,$1026,$1021,$1030,$0015,$0015,$0015,$0015,$1230,$1021,$5077,$5076,$5077,$5076,$1077,$1076,$1077,$1076,$1024,$0025,$0025,$1026,$1021,$1077,$1076,$1077,$1076,$1013,$0014,$0015,$1016,$1021,$1021,$1021,$1021
         dw $1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1013,$0014,$0015,$1016,$1021,$1021,$1021,$1024,$0025,$0025,$0025,$0025,$1026,$1021,$1021,$1021,$1013,$0014,$0015,$1016,$1030,$0015,$0015,$0015,$0015,$0015,$0015,$0015,$0015,$1033,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1013,$0014,$0015,$1016,$1021,$1024,$0025,$0025,$0025,$0025,$0025,$0025,$1026,$1021
         dw $1030,$0015,$0015,$0015,$0015,$1033,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1024,$0025,$0025,$1026,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1024,$0025,$0025,$0025,$0025,$1026,$1021,$1021,$1021,$1021,$1021,$1021,$1021
         dw $1030,$0015,$0015,$0015,$0015,$0015,$0015,$0015,$0015,$1033,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1024,$0025,$0025,$0025,$0025,$0025,$0025,$1026,$1021,$1030,$0015,$0015,$0015,$0015,$1033,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1024,$0025,$0025,$1026,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021,$1021

demos/zelda/App.Main.s

@@ -184,7 +184,7 @@ EvtLoop
                     sec
                     sbc        ScreenY1
                     cmp        #8
-                    bcs        :vsyncloop
+                    bcs        :vsyncloop             ; Wait until we're within the top 8 scanlines
                     lda        #1
                     jsl        SetBorderColor
 :no_vsync

src/Sprite.s

@@ -122,7 +122,12 @@ _ClearSpriteFromTileStore
 
 ; This function looks at the sprite list and renders the sprite plane data into the appropriate
 ; tiles in the code field.  There are a few phases to this routine.  The assumption is that
-; any sprite that needs to be re-drawn has been marked as DIRTY.
+; any sprite that needs to be re-drawn has been marked as DIRTY or DAMAGED.
+;
+; A DIRTY sprite is one that has moved, so it needs to be erased/redrawn in the sprite
+; buffer AND the tiles it covers marked for refresh.  A DAMAGED sprite shared one or more
+; tiles with a DIRTY sprite, so it needs to be redraw in the sprite buffer (but not erased!)
+; and its tile do NOT need to be marked for refresh.
 ;
 ; In the first phase, we run through the list of dirty sprites and erase them from their
 ; OLD_VBUFF_ADDR.  This clears the sprite plane buffers.  We also iterate through the
@@ -135,7 +140,6 @@ _ClearSpriteFromTileStore
 ; In the second phase, the sprite is re-drawn into the sprite plane buffers and the appropriate
 ; Tile Store locations are marked as dirty. It is important to recognize that the sprites themselves
 ; can be marked dirty, and the underlying tiles in the tile store are independently marked dirty.
-
 forceSpriteFlag ds 2
 _RenderSprites
 
@@ -143,7 +147,7 @@ _RenderSprites
 ; last time that the frame was rendered, then we need to mark all of the sprites as dirty so that
 ; the tiles on which they were located at the previous frame will be refreshed
 ;
-; OPTIMIZATION NOTE: Shoud check that the sprite actually chanegs position.  If the screen scrolles
+; OPTIMIZATION NOTE: Should check that the sprite actually changes position.  If the screen scrolls
 ;                    by +X, but the sprite moves by -X (so it's relative position is unchanged), then
 ;                    it does NOT need to be marked as dirty.
 
@@ -161,6 +165,32 @@ _RenderSprites
             sta   forceSpriteFlag
 :no_chng_y
 
+; Alter first phase. _OpenListHead is, essentially, a count of how many sprites. We can use that as an early-out
+; test to stop scanning the SPRITE_STATUS values once all active sprites have been accounted for.
+;            lda   _OpenListHead
+;            beq   :exit1
+;            lsr
+;            sta   tmp0
+
+;            lda   _Sprites+SPRITE_STATUS
+;            beq   :exit1
+;            ldy   #0
+;            jsr   _ClearSprite
+;            dec   tmp0
+;            beq   :exit1
+
+;            lda   _Sprites+SPRITE_STATUS+2
+;            beq   :exit1
+;            ldy   #2
+;            jsr   _ClearSprite
+;            dec   tmp0
+;            beq   :exit1
+
+;            ...
+
+;:exit1
+
+            
 ; First phase, erase all dirty sprites
             ldy   #0
 :loop1      lda   _Sprites+SPRITE_STATUS,y       ; If the status is zero, that's the sentinel value
@@ -281,37 +311,16 @@ _GetTileAt
             clc
             rts
 
-; X = _Sprites array offset
+; Y = _Sprites array offset
-_EraseSprite
-             txy
 _EraseSpriteY
              lda   _Sprites+OLD_VBUFF_ADDR,y
              beq   :noerase
-             lda   _Sprites+SPRITE_ID,y
+             ldx   _Sprites+SPRITE_DISP,y              ; get the dispatch index for this sprite
-             and   #$1800                        ; use bits 11 and 12 to dispatch (oly care about size)
+             jmp   (:do_erase,x)
-             lsr
-             lsr
-             xba
-             tax
-             jmp   (:erase_sprite,x)
 :noerase     rts
-:erase_sprite dw   erase_8x8,erase_8x16,erase_16x8,erase_16x16
+:do_erase    dw    _EraseTileSprite8x8,_EraseTileSprite8x16
+             dw    _EraseTileSprite16x8,_EraseTileSprite16x16
 
-erase_8x8
-             ldx   _Sprites+OLD_VBUFF_ADDR,y
-             jmp   _EraseTileSprite8x8                 ; erase from the old position
-
-erase_8x16
-             ldx   _Sprites+OLD_VBUFF_ADDR,y
-             jmp   _EraseTileSprite8x16
-
-erase_16x8
-             ldx   _Sprites+OLD_VBUFF_ADDR,y
-             jmp   _EraseTileSprite16x8
-
-erase_16x16
-             ldx   _Sprites+OLD_VBUFF_ADDR,y
-             jmp   _EraseTileSprite16x16
 
 ; X = _Sprites array offset
 _DrawSprite
@@ -768,8 +777,9 @@ _CacheSpriteBanks
 
 SPRITE_PLANE_SPAN equ 256
 
-; X = bank address
+; A = bank address
 _EraseTileSprite8x8
+            tax
             phb                                   ; Save the bank to switch to the sprite plane
 
             pei    SpriteBanks
@@ -795,6 +805,7 @@ _EraseTileSprite8x8
             rts
 
 _EraseTileSprite8x16
+            tax
             phb                                   ; Save the bank to switch to the sprite plane
 
             pei    SpriteBanks
@@ -820,6 +831,7 @@ _EraseTileSprite8x16
             rts
 
 _EraseTileSprite16x8
+            tax
             phb                                   ; Save the bank to switch to the sprite plane
 
             pei    SpriteBanks
@@ -849,6 +861,7 @@ _EraseTileSprite16x8
             rts
 
 _EraseTileSprite16x16
+            tax
             phb                                   ; Save the bank to switch to the sprite plane
 
             pei    SpriteBanks
@@ -968,6 +981,8 @@ _AddSprite
             jsr   _GetSpriteVBuffAddrTmp        ; Preserves X-register
             sta   _Sprites+VBUFF_ADDR,x
 
+            jsr   _PrecalcAllSpriteInfo         ; Cache stuff
+
             txa                                 ; And return the sprite ID
             clc                                 ; Mark that the sprite was successfully added
 
@@ -983,6 +998,22 @@ _AddSprite
 
             rts
 
+; Precalculate some cached values for a sprite.  These are *only* to make other part of code,
+; specifically the draw/erase routines more efficient.
+;
+; There are variations of thi routine based on whether we are adding a new sprite, updating
+; it's tile information, or changing its position.
+;
+; X = sprite index
+_PrecalcAllSpriteInfo
+            lda   _Sprites+SPRITE_ID,x 
+            and   #$1800                        ; use bits 11 and 12 to dispatch (oly care about size)
+            lsr
+            lsr
+            xba
+            sta   _Sprites+SPRITE_DISP,x
+            rts
+
 ; Remove a sprite from the list. Just mark its STATUS as FREE and it will be
 ; picked up in the next AddSprite.
 ;
@@ -1032,6 +1063,8 @@ _UpdateSpriteXnc
             jsr   _GetTileAddr                  ; This applies the TILE_ID_MASK
             sta   _Sprites+TILE_DATA_OFFSET,x
 
+            jsr   _PrecalcAllSpriteInfo         ; Cache stuff
+
             lda   #SPRITE_STATUS_DIRTY          ; Content is changing, mark as dirty
             sta   _Sprites+SPRITE_STATUS,x
 
@@ -1093,30 +1126,32 @@ _MoveSpriteXnc
 NUM_BUFF_LINES equ 24
 
 MAX_SPRITES  equ 16
-SPRITE_REC_SIZE equ 34
+SPRITE_REC_SIZE equ 36
 
-SPRITE_STATUS_EMPTY equ 0
+SPRITE_STATUS_EMPTY   equ 0         ; slot is unitialized / free
-SPRITE_STATUS_CLEAN equ 1
+SPRITE_STATUS_CLEAN   equ 1         ; 
-SPRITE_STATUS_DIRTY equ 2
+SPRITE_STATUS_DIRTY   equ 2
-SPRITE_STATUS_FREE  equ 4
+SPRITE_STATUS_FREE    equ 4
+SPRITE_STATUS_DAMAGED equ 8
 
-SPRITE_STATUS equ {MAX_SPRITES*0}
+SPRITE_STATUS      equ {MAX_SPRITES*0}
-TILE_DATA_OFFSET equ {MAX_SPRITES*2}
+TILE_DATA_OFFSET   equ {MAX_SPRITES*2}
-VBUFF_ADDR equ {MAX_SPRITES*4}
+VBUFF_ADDR         equ {MAX_SPRITES*4}
-SPRITE_ID equ {MAX_SPRITES*6}
+SPRITE_ID          equ {MAX_SPRITES*6}
-SPRITE_X equ {MAX_SPRITES*8}
+SPRITE_X           equ {MAX_SPRITES*8}
-SPRITE_Y equ {MAX_SPRITES*10}
+SPRITE_Y           equ {MAX_SPRITES*10}
-OLD_VBUFF_ADDR equ {MAX_SPRITES*12}
+OLD_VBUFF_ADDR     equ {MAX_SPRITES*12}
-TILE_STORE_ADDR_1 equ {MAX_SPRITES*14}
+TILE_STORE_ADDR_1  equ {MAX_SPRITES*14}
-TILE_STORE_ADDR_2 equ {MAX_SPRITES*16}
+TILE_STORE_ADDR_2  equ {MAX_SPRITES*16}
-TILE_STORE_ADDR_3 equ {MAX_SPRITES*18}
+TILE_STORE_ADDR_3  equ {MAX_SPRITES*18}
-TILE_STORE_ADDR_4 equ {MAX_SPRITES*20}
+TILE_STORE_ADDR_4  equ {MAX_SPRITES*20}
-TILE_STORE_ADDR_5 equ {MAX_SPRITES*22}
+TILE_STORE_ADDR_5  equ {MAX_SPRITES*22}
-TILE_STORE_ADDR_6 equ {MAX_SPRITES*24}
+TILE_STORE_ADDR_6  equ {MAX_SPRITES*24}
-TILE_STORE_ADDR_7 equ {MAX_SPRITES*26}
+TILE_STORE_ADDR_7  equ {MAX_SPRITES*26}
-TILE_STORE_ADDR_8 equ {MAX_SPRITES*28}
+TILE_STORE_ADDR_8  equ {MAX_SPRITES*28}
-TILE_STORE_ADDR_9 equ {MAX_SPRITES*30}
+TILE_STORE_ADDR_9  equ {MAX_SPRITES*30}
 TILE_STORE_ADDR_10 equ {MAX_SPRITES*32}
+SPRITE_DISP        equ {MAX_SPRITES*34}  ; pre-calculated index for jmp (abs,x) based on sprite size
 
 ; Maintain the index of the next open sprite slot.  This allows us to have amortized
 ; constant sprite add performance.  A negative value means no slots are available.

src/Sprite2.s

@@ -84,21 +84,18 @@ mdsOut  rts
 _MarkDirtySprite
 
         stz   _Sprites+TILE_STORE_ADDR_1,x       ; Clear the this sprite's dirty tile list in case of an early exit
-        lda   _SpriteBits,x
+        lda   _SpriteBits,x                      ; Cache its bit flag to mark in the tile slots
         sta   SpriteBit
 
 ; Clip the sprite's extent to the screen so we can assume (mostly) position values from here on out.  Note that
 ; the sprite width and height are _only_ used in the clip and afterward all calculation use the clip rect
 ;
 ; OPTIMIZATION NODE: These values can be calculated in AddSprite/MoveSprite once and stored in the sprite
-;                    record since the screen size doesn't change.
+;                    record since the screen size doesn't change.  An off-screen flag can be set.
 
-        lda   _Sprites+SPRITE_ID,x               ; Get an index into the height/width tables based on the sprite bits
+        ldy   _Sprites+SPRITE_DISP,x             ; Get an index into the height/width tables based on the sprite bits
-        and   #$1800
+;        lda   _Sprites+IS_OFF_SCREEN,x           ; Check if the sprite is visible in the playfield
-        xba
+;        bne   mdsOut
-        lsr
-        lsr
-        tay
 
         lda   _Sprites+SPRITE_X,x
         bpl   :pos_x
@@ -189,7 +186,8 @@ _MarkDirtySprite
         sta   ColLeft
 
 ; Sneak a pre-calculation here. Calculate the upper-left corder of the sprite in the sprite plane.
-; We can reuse this in all of the routines below
+; We can reuse this in all of the routines below.  This is not the (x,y) of the sprite itself, but
+; the corner of the tile it overlaps with
 
         clc
         lda   TileTop
@@ -219,10 +217,10 @@ _MarkDirtySprite
 ; Dispatch to the calculated sizing
 
 ; Begin a list of subroutines to cover all of the valid sprite size compinations.  This is all unrolled code,
-; maily to be able to do an unrolled fill of the TILE_STORE_ADDR_X values.  Thus, it's important that the clipping
+; mainly to be able to do an unrolled fill of the TILE_STORE_ADDR_X values.  Thus, it's important that the clipping
-; function does its job properly since it allows up to save a lot of time here.
+; function does its job properly since it allows us to save a lot of time here.
 ;
-; These functional are a trade off of being composable versus fast.  Having to pay for multiple JSR/RTS invoations
+; These functions are a trade off of being composable versus fast.  Having to pay for multiple JSR/RTS invocations
 ; in the hot sprite path isn't great, but we're at a point of diminishing returns.
 ;
 ; There *might* be some speed gained by pushing a list of :mark_R_C addressed onto the stack in the clipping routing

src/blitter/Tiles.s

@@ -678,10 +678,10 @@ _PushDirtyTileX
                  lda  TileStore+TS_DIRTY,x
                  bpl  :occupied2
 
-                 lda  DirtyTileCount
+                 txa                                  ; any non-negative value will work, this saves work below
-                 sta  TileStore+TS_DIRTY,x
+                 sta  TileStore+TS_DIRTY,x            ; and is 1 cycle fater than loading a constanct value
 
-                 txa
+;                 txa
                  ldx  DirtyTileCount
                  sta  DirtyTiles,x