GPU instruction pair <= 148 cycles:
        setup
        (burst)
        clean

Pixel burst (videoA,B and/or C)

                CopperSetup
    Burst Burst Burst Burst
    CopperClean CopperSetup
    Burst Burst Burst Burst
    CopperClean

Setup:
        ; Overhead
        ld [videoY],x   ; Find out page from video indirection table
        ld $01,y
        ls [y,x]
        st [vTmp],y

        ; Object
        ld [y,$a0]      ; X position of object
        ld ac,x

        ; Pixel 1
        ld [y,x]        ; Preserve background
        st [y,$a1]
        ld [y,$a2]      ; Display object
        st [y,x++]

Clean:
        ; Overhead
        ld [vTmp],y     ; XXX is this untouched by the video loop?

        ld [y,$a0]      ; X position of object
        ld ac,x
        ld [y,$a1]      ; Restore background
        st [y,x++]

Object costs:   4 cycles
Pixel costs:    6 cycles

5 pixel object: 34 cycles -> 4 objects (ghosts!), 10 cycles for overhead sounds ok
6 pixel object: 40 cycles -> 3 objects (boring)

Eg. 4 objects of 6 pixels

148/4 = 37

 - # # # - - - - - -
 # O # O # - - - - -
 # # # # # - - X - -
 # # # # # - - - - -
 # - # - # - - - - -

 - O O O -
 O O # O O
 O O O O O
 O O O O O
 - O O O -

 - O O O -
 O O # O O
 - O O O O
 O O O O O
 - O O O -

 O O O O -
 - O # O O
 - - O O O
 - O O O O
 O O O O -

Because most lines have no sprites, there is a lot of wasted CPU time.
These kind of sprites are better done in vertical blank.


But here's an idea for a cool screensaver:

Setup:
        ld [y,$a7]
        st [vTmp],x
        ld [y,$a8]
        adda sprite1
        bra [ac]
sprite1 ld [y,$a0]      ;Draw sprite 1
        st [y,x++]
        ld [y,$a1]
        st [y,x++]
        ld [y,$a2]
        st [y,x++]
        ld [y,$a3]
        st [y,x++]
        ld [y,$a4]
        st [y,x++]
        ld [y,$a5]
        st [y,x++]
        ld [y,$a6]
        st [y,x++]

        ld [y,$a0]
loop    bne loop        ;Resync
        suba 2

Clean:
        ld [vTmp],x

        ld [x]          ;Restore sprite 1
        st [y,x++]
        ld [x]
        st [y,x++]
        ld [x]
        st [y,x++]
        ld [x]
        st [y,x++]
        ld [x]
        st [y,x++]
        ld [x]
        st [y,x++]
        ld [x]
        st [y,x++]

With a background in 160 bytes in zero page, we have a vertical bars.
Playing with videoTable[].dX gives us a blocky pattern!

7 pixel objects cost 8*4 = 32 cycles. We can have 4 objects <-- lets try this
6 pixel objects cost 7*4 = 28 cycles. We can have 5 objects

But no alpha channel..
Different ROM routine per shape?

        ld [y,x]
        op [y,$a0]
        st [y,x++]

Maybe we don't need full screen frame buffer. 1 line is enough?